1 00:00:00,050 --> 00:00:02,720 curtis: I'm pretty sure we've said smoking hole more times 2 00:00:02,720 --> 00:00:05,000 than we've on this podcast. 3 00:00:05,030 --> 00:00:05,840 Just for the record. 4 00:00:06,410 --> 00:00:07,010 Just saying 5 00:00:08,389 --> 00:00:08,809 It's getting a 6 00:00:08,809 --> 00:00:09,680 lot of play today. 7 00:00:29,260 --> 00:00:32,140 Hi and welcome to Backup Central's Restore it All podcast. 8 00:00:32,140 --> 00:00:32,840 I'm your host, W. 9 00:00:32,840 --> 00:00:34,540 Curtis Preston, AKA Mr. 10 00:00:34,540 --> 00:00:39,110 Backup and I have with me, my Bhangra dance consultant, Prasanna 11 00:00:39,110 --> 00:00:40,960 Malaiyandi, how's it going Prasanna? 12 00:00:41,550 --> 00:00:43,200 Prasanna: I'm good Curtis, but I have to warn you. 13 00:00:43,200 --> 00:00:44,970 I have not a dancer at all. 14 00:00:44,970 --> 00:00:49,800 So probably the wrong person to be seeking advice about dancing from, 15 00:00:50,470 --> 00:00:53,680 curtis: But you said that you knew about Bhangra dancing and that you 16 00:00:53,680 --> 00:00:55,120 could advise me on these things. 17 00:00:55,950 --> 00:01:02,000 Prasanna: I told you that it's like a Indian dance style, if you will. 18 00:01:02,000 --> 00:01:05,540 And you had asked a question of, have I seen it because I bet I've 19 00:01:05,540 --> 00:01:06,950 seen a bunch of Bollywood movies. 20 00:01:07,050 --> 00:01:12,000 curtis: You expanded my horizon I bought my tickets my wife and I will be going 21 00:01:12,000 --> 00:01:15,230 to see the show it's called Bhangin' it! 22 00:01:16,545 --> 00:01:22,335 It's bangin' was spelled would be H so it's it's trying to like 23 00:01:23,235 --> 00:01:25,085 do an homage to the Bhangra. 24 00:01:26,175 --> 00:01:33,375 So it's a, new musical at the LA Jolla Playhouse, which is a very nice 25 00:01:33,705 --> 00:01:35,025 Playhouse that I've actually never been. 26 00:01:35,025 --> 00:01:36,195 I've lived here 20 something years. 27 00:01:36,195 --> 00:01:39,435 I've never watched a show there, but a lot of like big Broadway 28 00:01:39,435 --> 00:01:40,665 shows actually start out. 29 00:01:40,725 --> 00:01:41,355 I've never started. 30 00:01:42,135 --> 00:01:43,815 I've, always watched the Broadway shows 31 00:01:44,590 --> 00:01:44,960 Prasanna: Like 32 00:01:46,350 --> 00:01:47,070 gone to Broadway. 33 00:01:47,070 --> 00:01:50,040 curtis: This is the kind of show that could possibly hit big on Broadway. 34 00:01:50,460 --> 00:01:55,620 And so we'll see it and we'll see if it's any good and 35 00:01:55,850 --> 00:01:56,170 I'll 36 00:01:56,285 --> 00:01:56,795 Prasanna: waiting for 37 00:01:56,880 --> 00:01:57,570 curtis: with my review. 38 00:01:58,115 --> 00:01:58,595 Prasanna: Yes. 39 00:01:59,495 --> 00:02:01,265 I think our listeners will be curious. 40 00:02:02,405 --> 00:02:04,925 And by the way, for those in San Diego, when is it running? 41 00:02:05,895 --> 00:02:06,685 Do you know how long? 42 00:02:06,735 --> 00:02:07,545 curtis: It's running. 43 00:02:07,595 --> 00:02:08,675 It's running until April. 44 00:02:09,315 --> 00:02:09,825 Prasanna: Okay. 45 00:02:10,485 --> 00:02:10,785 So that 46 00:02:10,785 --> 00:02:10,905 was 47 00:02:10,905 --> 00:02:11,885 folks in San Diego. 48 00:02:12,265 --> 00:02:12,715 curtis: Yeah. 49 00:02:14,505 --> 00:02:20,625 Yeah, depending on when this goes live, if it goes live less than a month from 50 00:02:20,625 --> 00:02:27,045 now, then you have two days left to go see it because it runs until April 51 00:02:27,045 --> 00:02:31,395 17th at the LA Jolla Playhouse, by which time all the tickets will be 52 00:02:31,395 --> 00:02:32,685 gone and you won't be able to see it. 53 00:02:33,645 --> 00:02:40,245 Sorry, I don't know what to tell you, but so we, have a longtime 54 00:02:40,245 --> 00:02:42,105 friend on the podcast here today. 55 00:02:42,105 --> 00:02:42,665 Prasanna. 56 00:02:43,275 --> 00:02:47,295 I'm excited to bring him on I, and not just because he's one of those 57 00:02:47,295 --> 00:02:48,675 people that make me feel young. 58 00:02:52,515 --> 00:02:57,105 As, been in the it industry for an awfully long time, makes me feel like 59 00:02:57,135 --> 00:02:58,875 a young whippersnapper sometimes. 60 00:02:59,245 --> 00:03:05,635 He is now the technologist extraordinary and plenipotentiary at Vast Data. 61 00:03:05,885 --> 00:03:08,715 Welcome to the podcast, Howard Marks. 62 00:03:09,650 --> 00:03:10,070 Howard: Thank you. 63 00:03:10,070 --> 00:03:11,300 It's very nice to be here. 64 00:03:11,990 --> 00:03:15,830 I was always about Fauci guy, so I don't know much about Indian dance. 65 00:03:16,485 --> 00:03:18,225 Prasanna: Curtis didn't either before he met me. 66 00:03:18,225 --> 00:03:19,005 So it's fine. 67 00:03:20,555 --> 00:03:31,115 curtis: Yeah I, my knowledge of Indian dance it basically includes the 68 00:03:31,115 --> 00:03:34,145 reference to it in what was that movie? 69 00:03:35,580 --> 00:03:36,090 Prasanna: Millionaire. 70 00:03:36,665 --> 00:03:39,305 curtis: bright, the BR the bride and prejudice. 71 00:03:39,900 --> 00:03:40,230 Prasanna: Oh 72 00:03:40,385 --> 00:03:40,835 curtis: There's a 73 00:03:41,580 --> 00:03:41,760 Prasanna: yeah. 74 00:03:41,760 --> 00:03:42,360 I th I think 75 00:03:42,695 --> 00:03:46,805 curtis: there's a, it's a pride and prejudice, 76 00:03:47,010 --> 00:03:47,340 Prasanna: yeah. 77 00:03:48,305 --> 00:03:51,725 curtis: Knock off done with what's her name? 78 00:03:51,895 --> 00:03:52,680 . Prasanna: Ashwaryia Rai. 79 00:03:52,705 --> 00:03:53,145 I think. 80 00:03:54,065 --> 00:03:58,505 curtis: She remember she, she, in the movie she, gives two D two dance moves. 81 00:03:58,505 --> 00:04:01,145 It was petting the dog and screwing in the light bulb. 82 00:04:01,265 --> 00:04:02,285 I don't know if you remember that. 83 00:04:02,285 --> 00:04:03,095 She says that. 84 00:04:06,375 --> 00:04:08,895 That's literally the extent of my knowledge of Indian dance. 85 00:04:09,015 --> 00:04:11,235 That, and the fact that I've watched a bunch of Bollywood movies, but 86 00:04:11,745 --> 00:04:13,125 that's all thanks to Prasanna. 87 00:04:13,545 --> 00:04:13,755 Prasanna: Yeah. 88 00:04:15,615 --> 00:04:18,405 curtis: So you never know what you're going to get when you're listening to the 89 00:04:18,945 --> 00:04:20,805 Backup Central Restore it All podcast. 90 00:04:22,275 --> 00:04:25,155 Speaking of which, let me throw out our usual disclaimer, Prasanna 91 00:04:25,155 --> 00:04:26,405 and I work for different companies. 92 00:04:26,405 --> 00:04:27,425 Persona works for Zoom. 93 00:04:27,425 --> 00:04:28,145 I worked for Druva. 94 00:04:28,175 --> 00:04:31,235 This is not a podcast of either company and the opinions that you hear are 95 00:04:31,295 --> 00:04:37,565 ours, and be sure to rate this podcast ratethispodcast.com/restore, or just 96 00:04:37,565 --> 00:04:43,655 go at your on your favorite pod catcher apple podcasts and just scroll down 97 00:04:43,655 --> 00:04:45,605 to the bottom and give us some stars. 98 00:04:45,605 --> 00:04:49,025 And if you really want to make my day, actually put some words there. 99 00:04:49,505 --> 00:04:50,705 Yeah, absolutely. 100 00:04:51,155 --> 00:04:55,175 And if you are interested in the things that we're interested in, like 101 00:04:55,175 --> 00:05:00,250 backups and storage and resilience and ransomware recovery and cyber 102 00:05:00,250 --> 00:05:02,110 warfare and all of these things. 103 00:05:03,430 --> 00:05:09,940 Then just send me a note @wcpreston on Twitter, or wcurtispreston@gmail, and 104 00:05:10,040 --> 00:05:13,870 I'll be happy to get you on the podcast. 105 00:05:14,140 --> 00:05:14,710 Prasanna: friendly. 106 00:05:14,860 --> 00:05:16,240 We ask questions. 107 00:05:17,050 --> 00:05:21,580 curtis: we even apparently, although the last episode I said, unless your 108 00:05:21,580 --> 00:05:25,160 name was Stewart and apparently Stewart has now reached out to you Prasanna 109 00:05:25,320 --> 00:05:26,190 Prasanna: Yes, he has. 110 00:05:26,520 --> 00:05:26,700 He 111 00:05:26,900 --> 00:05:27,380 curtis: and, 112 00:05:30,600 --> 00:05:31,170 Howard: So even 113 00:05:31,250 --> 00:05:31,370 curtis: and 114 00:05:31,420 --> 00:05:31,690 Howard: name 115 00:05:31,720 --> 00:05:34,510 curtis: Even Stuart can get on this podcast. 116 00:05:34,510 --> 00:05:42,010 So if we're going to let you know a mouse on the podcast, then surely we can let 117 00:05:42,010 --> 00:05:44,770 you, his name is Stuart Liddle for those of you that didn't get that reference 118 00:05:44,770 --> 00:05:45,100 anyway. 119 00:05:45,300 --> 00:05:46,920 Howard: to make me feel honored here. 120 00:05:50,975 --> 00:05:52,780 curtis: We literally let anybody in the door, 121 00:05:55,990 --> 00:05:59,140 including guys who always wear Hawaiian shirts. 122 00:06:01,235 --> 00:06:01,985 Howard: They're comfortable. 123 00:06:01,985 --> 00:06:07,415 They come in my size and at this point I'm just known for them. 124 00:06:07,835 --> 00:06:11,945 I have been known to tell people I'm going to meet at the 125 00:06:11,945 --> 00:06:13,685 Starbucks at some conference. 126 00:06:14,315 --> 00:06:16,565 Just look for Santa Clause in an Aloha shirt. 127 00:06:16,565 --> 00:06:17,495 That will be me. 128 00:06:21,160 --> 00:06:21,310 curtis: much. 129 00:06:21,310 --> 00:06:21,940 It pretty much 130 00:06:22,250 --> 00:06:22,550 Prasanna: that's. 131 00:06:24,205 --> 00:06:24,475 Howard: Yeah. 132 00:06:24,745 --> 00:06:29,545 You know how many 350 pound guys with a gray beard are there walking around the 133 00:06:29,545 --> 00:06:32,485 average tech show, wearing an Aloha shirt? 134 00:06:32,835 --> 00:06:33,495 Two 135 00:06:33,680 --> 00:06:35,210 curtis: I'm going to, yeah. 136 00:06:35,270 --> 00:06:36,110 Two, yeah. 137 00:06:36,500 --> 00:06:37,070 At most. 138 00:06:37,100 --> 00:06:37,820 Absolutely. 139 00:06:38,420 --> 00:06:39,620 And one of them is going to be you. 140 00:06:40,400 --> 00:06:40,555 Howard: Yeah. 141 00:06:41,670 --> 00:06:45,050 curtis: so how long have you been at Vast Data? 142 00:06:45,745 --> 00:06:49,705 Howard: I've been at Vast Data three years and 15 days. 143 00:06:50,545 --> 00:06:51,175 curtis: Wow. 144 00:06:52,755 --> 00:06:54,975 Prasanna: And the company is fairly new as well. 145 00:06:55,795 --> 00:06:58,915 Howard: I joined Vast Data the day before we came out of stealth. 146 00:06:59,845 --> 00:07:06,835 My, my first official act at Vast Data was a briefing for Chris Mellor followed 147 00:07:06,835 --> 00:07:08,755 the next day by Storage Field Day. 148 00:07:09,955 --> 00:07:10,525 curtis: Wow. 149 00:07:11,155 --> 00:07:13,015 Howard: Nothing like starting off running 150 00:07:15,525 --> 00:07:19,945 Now, I joined Vast from being an independent analyst. 151 00:07:20,275 --> 00:07:25,315 So there were a couple of weeks there where I was getting brought up to speed 152 00:07:25,345 --> 00:07:28,315 and such before my official start date. 153 00:07:30,445 --> 00:07:30,895 But yeah, 154 00:07:30,945 --> 00:07:36,715 curtis: And why don't you give a for those that aren't familiar with Vast 155 00:07:36,735 --> 00:07:39,995 Data, give us a, know, the elevator 156 00:07:40,235 --> 00:07:40,385 Howard: sure. 157 00:07:40,385 --> 00:07:40,715 curtis: and 158 00:07:41,240 --> 00:07:47,960 Howard: The really short form on Vast Data is that we make very large scale all 159 00:07:47,960 --> 00:07:51,500 flash file and object storage systems. 160 00:07:52,070 --> 00:07:58,640 And when I say very large scale our average selling price for 161 00:07:58,640 --> 00:08:02,510 our cluster is well on the north side of a million dollars. 162 00:08:02,510 --> 00:08:04,760 It's multiple petabytes. 163 00:08:05,600 --> 00:08:12,500 Today we're just introducing a new storage enclosure that brings 164 00:08:12,500 --> 00:08:22,940 our building block down from 675 terabytes per HA enclosure to 338. 165 00:08:22,940 --> 00:08:24,950 So we're taking it down by factor of two. 166 00:08:24,950 --> 00:08:28,100 We're going from a two U to a one U enclosure. 167 00:08:28,370 --> 00:08:34,250 We'll talk about that in a little bit, but the innovative thing 168 00:08:34,250 --> 00:08:37,160 about Vast is the architecture. 169 00:08:37,610 --> 00:08:43,520 If you talk about a large scale system, like we build traditionally, that's been 170 00:08:43,520 --> 00:08:50,960 done with a scale out, shared nothing model where you have a lot of x86 servers. 171 00:08:51,230 --> 00:08:58,070 Each of those x86 servers owns some set of media and they communicate 172 00:08:58,070 --> 00:09:01,580 on a backend network and software makes it look like one big system. 173 00:09:03,320 --> 00:09:06,710 But those systems start to break down at really large scale. 174 00:09:07,430 --> 00:09:09,410 And so we've come up with a new model. 175 00:09:09,410 --> 00:09:19,340 We call DASE the shared everything architecture instead of having a field of 176 00:09:19,370 --> 00:09:27,020 peer nodes, each of which owns some media, we disaggregated the media into these HA 177 00:09:27,020 --> 00:09:29,270 enclosures that I was just talking about. 178 00:09:29,660 --> 00:09:38,755 So no single point of failure, 400 gig connections to an NVME fabric and 179 00:09:38,785 --> 00:09:40,735 that's typically a hundred gig Ethernet. 180 00:09:40,975 --> 00:09:44,125 Some of our HPC customers like to run InfiniBand so we 181 00:09:44,125 --> 00:09:45,565 can do InfiniBand as well. 182 00:09:47,485 --> 00:09:51,115 All those enclosures do is hold data. 183 00:09:51,385 --> 00:09:53,275 There's no services there. 184 00:09:54,475 --> 00:09:59,975 All of the services, everything that you would think of as the controller function 185 00:10:00,575 --> 00:10:07,475 of the system runs in stateless Docker containers in the front end servers. 186 00:10:08,975 --> 00:10:13,055 So when a user makes a request to a protocol server to one of 187 00:10:13,055 --> 00:10:18,095 those front end servers could be NFS, could be SMB, could be S3. 188 00:10:18,785 --> 00:10:25,115 That server looks in the metadata that's stored in storage class memory 189 00:10:25,115 --> 00:10:32,375 in the enclosures, finds the data the user's requesting in the data in 190 00:10:32,375 --> 00:10:38,685 QLC flash in those same enclosures, retrieves it over the NVME over fabric's 191 00:10:38,705 --> 00:10:41,555 fabric and delivers it to the user. 192 00:10:41,975 --> 00:10:50,015 So there's none of the traffic from node to node required to reassemble 193 00:10:50,015 --> 00:10:56,585 data, everything's north, south across that NVME over fabrics connection. 194 00:10:57,035 --> 00:11:03,485 And since the metadata is in storage class memory, it's fast enough to 195 00:11:03,485 --> 00:11:09,515 directly access by all of the front end servers that they can just share it. 196 00:11:09,545 --> 00:11:10,745 They don't have to cash it. 197 00:11:11,765 --> 00:11:15,395 And by not having the cache, we don't have all the complexities 198 00:11:15,395 --> 00:11:17,035 of keeping the cache coherent. 199 00:11:17,475 --> 00:11:19,005 Prasanna: I was just going to ask about that, Howard. 200 00:11:19,005 --> 00:11:23,685 So it looks like though you're dis-aggregating the actual storage 201 00:11:23,685 --> 00:11:27,735 and metadata from all the front end processing, which allows, 202 00:11:28,245 --> 00:11:31,695 would assume the front end to scale independently of the backend. 203 00:11:32,985 --> 00:11:37,335 Howard: So each of those front end protocol servers, mounts all of the 204 00:11:37,335 --> 00:11:39,825 SSDs in the cluster at boot time. 205 00:11:40,755 --> 00:11:47,445 And then it looks at all of those SSDs, and at those are the SCM 206 00:11:47,445 --> 00:11:51,705 SSDs that hold the metadata and the QLC SSDs that hold the data. 207 00:11:52,995 --> 00:11:55,845 So everybody has access to everything. 208 00:11:56,685 --> 00:12:01,335 And instead of sending messages back and forth between the front end servers, 209 00:12:01,875 --> 00:12:09,915 they simply write a single of truth in the shared metadata, so that the 210 00:12:09,915 --> 00:12:14,235 old so that you can place a lock on the metadata or update the metadata. 211 00:12:14,505 --> 00:12:18,405 But you never have to tell everybody else you updated it because if they want 212 00:12:18,405 --> 00:12:21,825 to know what the state is, they'll go look in the one place where it's true. 213 00:12:22,085 --> 00:12:22,295 Prasanna: Yeah. 214 00:12:22,325 --> 00:12:26,105 And because everything is stateless in the front end, you don't have to worry 215 00:12:26,105 --> 00:12:27,545 about that necessarily to everyone 216 00:12:27,655 --> 00:12:27,955 Howard: Right, 217 00:12:28,115 --> 00:12:28,805 Prasanna: that backend 218 00:12:29,425 --> 00:12:29,725 Howard: right. 219 00:12:30,615 --> 00:12:34,265 curtis: So the backend has both SSDs and QLC. 220 00:12:35,395 --> 00:12:40,735 Howard: What has SCM sort of storage class memory SSDs, and that can be 221 00:12:40,735 --> 00:12:47,875 Optane or and it has low end QLC SSDs. 222 00:12:48,510 --> 00:12:48,780 Prasanna: So 223 00:12:49,135 --> 00:12:55,765 curtis: And the, the, yeah the, storage class memory is what's 224 00:12:55,765 --> 00:12:57,535 holding the metadata and the 225 00:12:57,535 --> 00:12:59,035 QLC is, what's holding the data. 226 00:12:59,795 --> 00:13:00,455 Howard: Primarily. 227 00:13:00,455 --> 00:13:02,045 It's also used as a write buffer. 228 00:13:02,075 --> 00:13:02,495 curtis: Okay. 229 00:13:02,675 --> 00:13:03,095 Okay. 230 00:13:03,815 --> 00:13:10,655 Howard: So writes come into the storage class memory and get mirrored to two 231 00:13:10,655 --> 00:13:13,865 different SCM SSDs and then get ACKd. 232 00:13:14,975 --> 00:13:19,835 And then the migration from SCM to QLC happens after the act. 233 00:13:19,835 --> 00:13:22,445 So we have more time to do things like compress more fully. 234 00:13:22,582 --> 00:13:24,682 curtis: This is a very different game than. 235 00:13:26,447 --> 00:13:32,837 This idea of all of the front end nodes, being able to mount the entire 236 00:13:34,507 --> 00:13:35,047 Howard: Yes. 237 00:13:35,087 --> 00:13:35,867 curtis: the background 238 00:13:36,097 --> 00:13:36,307 Howard: Yeah. 239 00:13:36,307 --> 00:13:40,537 We we eliminate the whole concept of ownership and all the 240 00:13:40,537 --> 00:13:42,637 complexity that, that creates. 241 00:13:44,457 --> 00:13:49,247 And now I'm going to blow your mind because when I say the metadata is in 242 00:13:49,247 --> 00:13:54,257 the SCM, I don't mean just the element store metadata, the metadata for our 243 00:13:54,287 --> 00:13:59,967 merged file system object store, but also the data reduction metadata. 244 00:14:00,592 --> 00:14:07,132 And so when you add another enclosure to the cluster, you add more SCM, which 245 00:14:07,132 --> 00:14:09,712 means you add more room for that metadata. 246 00:14:10,132 --> 00:14:15,322 So regardless of the size of cluster, the cluster is one data reduction realm 247 00:14:15,652 --> 00:14:18,802 across tens or hundreds of petabytes. 248 00:14:18,827 --> 00:14:22,637 Prasanna: Because everything's looks like one cluster, if you will, or one system. 249 00:14:22,982 --> 00:14:23,462 Howard: right. 250 00:14:23,872 --> 00:14:27,272 And, we don't have to hold the data deduplication hash 251 00:14:27,272 --> 00:14:30,062 table in memory any place. 252 00:14:30,272 --> 00:14:33,722 It's all in SCM where it's fast enough we don't need that. 253 00:14:34,082 --> 00:14:38,312 So we don't have the limitations of how big a deduplication realm can be 254 00:14:38,882 --> 00:14:41,612 that most deduplication systems have. 255 00:14:42,887 --> 00:14:43,247 curtis: right. 256 00:14:43,247 --> 00:14:47,657 They typically top out around a a petabyte or so, and then you 257 00:14:47,657 --> 00:14:49,487 can't get any bigger than that. 258 00:14:50,237 --> 00:14:54,437 I don't know where to start on my questions! 259 00:14:55,537 --> 00:15:01,567 Howard: so from that, from the backup point of view, we're discovering that 260 00:15:02,797 --> 00:15:11,407 the customers are starting to demand higher restore speeds that traditionally 261 00:15:11,947 --> 00:15:15,997 all a customer worried about when they were picking the storage for their 262 00:15:15,997 --> 00:15:20,437 backups was it fast enough that I can make my backup within the window? 263 00:15:23,257 --> 00:15:29,437 And so we got systems like Data Domain and other disk based deduplicating systems, 264 00:15:30,337 --> 00:15:37,057 where there was a big write read asymmetry where you could write data faster to 265 00:15:37,057 --> 00:15:39,337 them than you could read data from them. 266 00:15:40,807 --> 00:15:46,597 Because reading data that caused the system to rehydrate turned 267 00:15:46,987 --> 00:15:49,377 sequential IO into random IO. 268 00:15:50,152 --> 00:15:52,612 And they had disks on the backend. 269 00:15:53,572 --> 00:15:57,262 And as disk drives have gotten bigger, this has gotten worse 270 00:15:58,042 --> 00:16:01,582 because a 20 terabyte disk drive today delivers exactly the same 271 00:16:01,582 --> 00:16:04,762 number of IOPS that a one terabyte disc drive delivered 10 years ago. 272 00:16:05,842 --> 00:16:11,362 So now 20 terabytes of data gets a 20th as many IOPS. 273 00:16:12,862 --> 00:16:17,662 And so you discover, yes, it takes me eight hours to back this up. 274 00:16:17,662 --> 00:16:19,762 It takes me 82 hours to restore it 275 00:16:21,692 --> 00:16:22,472 and 276 00:16:22,882 --> 00:16:23,062 curtis: Yeah. 277 00:16:23,122 --> 00:16:28,162 D D dedupe has never been very friendly for, large restores, especially if 278 00:16:28,162 --> 00:16:32,092 you're doing any sort of, if you want to do a live mount, forget it right. 279 00:16:32,122 --> 00:16:34,282 From a directly, from a Data Domain. 280 00:16:36,052 --> 00:16:39,772 It's possible in the same way, it's possible that... 281 00:16:39,772 --> 00:16:45,822 Howard: That's, but that's, you can bring up the Oracle or the SQL server VM. 282 00:16:46,687 --> 00:16:52,237 So that the it guys can access the passwords database, so that everybody 283 00:16:52,237 --> 00:16:55,207 can start at running ERP on it again. 284 00:16:55,237 --> 00:16:55,397 Prasanna: Yeah. 285 00:16:55,717 --> 00:16:56,887 Don't use it as production. 286 00:16:56,917 --> 00:16:57,667 That's a bad thing. 287 00:16:58,677 --> 00:16:58,867 Howard: Right. 288 00:16:58,947 --> 00:16:59,377 curtis: right. 289 00:17:00,187 --> 00:17:06,367 Howard: And we're discovering that people's requirements are getting tighter. 290 00:17:08,017 --> 00:17:14,467 You start thinking about software as a service providers where, you know, if you 291 00:17:14,467 --> 00:17:20,107 run some account, some industry specific accounting as a service for a thousand 292 00:17:20,107 --> 00:17:22,837 customers, that's a thousand databases. 293 00:17:23,797 --> 00:17:27,307 And when something goes wrong, you want to restore those databases 294 00:17:27,307 --> 00:17:31,297 as fast as you can, because your customers are going to be standing 295 00:17:31,357 --> 00:17:32,947 over your shoulder, yelling at you. 296 00:17:34,237 --> 00:17:39,097 And the last thing that's kicked, a couple of our potential customers over 297 00:17:39,097 --> 00:17:41,567 the edge is the ransomware threat. 298 00:17:43,217 --> 00:17:47,452 Because the size of the restore grows so much with ransomware. 299 00:17:48,862 --> 00:17:51,742 You start off with, they need to protect my data against ransomware 300 00:17:52,162 --> 00:17:54,322 and use various methods to do that. 301 00:17:54,322 --> 00:17:57,742 And so we have indestructable snapshots. 302 00:17:57,742 --> 00:18:05,002 So you can say snapshot this folder at 6:00 AM when the backup window 303 00:18:05,002 --> 00:18:08,152 closes and retain it for 30 days. 304 00:18:08,152 --> 00:18:11,272 And even if the administrator wants to delete it he can't. 305 00:18:11,817 --> 00:18:12,117 Prasanna: So I 306 00:18:12,142 --> 00:18:12,652 Howard: but 307 00:18:12,657 --> 00:18:13,197 Prasanna: about that. 308 00:18:14,847 --> 00:18:18,117 So I did read a little small blurb about that. 309 00:18:18,117 --> 00:18:18,567 So 310 00:18:20,127 --> 00:18:23,157 What prevents, is that locked down forever? 311 00:18:23,157 --> 00:18:27,147 Like an admin can't delete it no matter what, or is it just, there 312 00:18:27,147 --> 00:18:30,987 are additional safeguards in place to make sure that someone doesn't 313 00:18:30,987 --> 00:18:32,307 compromise the admin password, 314 00:18:32,392 --> 00:18:41,177 Howard: Anyone who ever talked to any customer of EMC Centera knows that if you 315 00:18:41,177 --> 00:18:48,167 build a system where you literally can't delete data someone will get themselves in 316 00:18:48,167 --> 00:18:55,847 trouble and fill it a hundred percent up with junk, and it will be a bad situation. 317 00:18:57,137 --> 00:19:03,317 So you have to provide some mechanism for overriding this because customers 318 00:19:03,317 --> 00:19:05,147 will paint themselves in corners. 319 00:19:07,727 --> 00:19:11,027 As I said, our average selling price is well over a million dollars. 320 00:19:11,027 --> 00:19:17,267 We don't have small customers who we only know third hand through VARs. 321 00:19:18,137 --> 00:19:22,787 We are in relatively intimate contact with every one of our customers. 322 00:19:23,327 --> 00:19:28,247 And so we don't have a fixed policy that says, if you jump through these 323 00:19:28,247 --> 00:19:34,097 hoops, then we will let you delete the undeletable snapshots we, and the 324 00:19:34,097 --> 00:19:36,277 customer agree what the hoops are. 325 00:19:36,992 --> 00:19:42,482 Yeah, multifactor authentication must be three of the five people on this list. 326 00:19:42,722 --> 00:19:48,032 They have to know the passphrase and the proper response to the passphrase. 327 00:19:48,302 --> 00:19:52,472 And if they respond with this other response to the passphrase, then for 328 00:19:52,472 --> 00:19:58,022 the next 24 hours, do not give anybody the secret as complicated as you want. 329 00:19:58,292 --> 00:20:01,472 We'll as long as we can write it down, those are the rules. 330 00:20:02,282 --> 00:20:08,432 And then once you've jumped through the hoops, we give you a time limited 331 00:20:08,462 --> 00:20:17,072 token that allows you to delete snapshots for a short period of time. 332 00:20:18,062 --> 00:20:21,712 And that token is a one-time pad. 333 00:20:23,177 --> 00:20:25,607 So that you can't re it's not good for 334 00:20:26,097 --> 00:20:26,307 Prasanna: Yeah. 335 00:20:26,837 --> 00:20:29,027 Howard: an hour whenever you use it. 336 00:20:29,447 --> 00:20:34,247 It is good for the time when we issue it for some limited period of time. 337 00:20:34,757 --> 00:20:36,767 And then you have to know the next one. 338 00:20:38,027 --> 00:20:42,407 And it's just, it was the best solution we could come up with. 339 00:20:43,327 --> 00:20:46,567 Prasanna: And this is probably helps in cases where someone 340 00:20:46,627 --> 00:20:50,467 attacks a company, they get access to the, to a storage system. 341 00:20:50,467 --> 00:20:54,097 They start deleting back-ups or what have you, it gives you 342 00:20:54,097 --> 00:20:55,747 that extra layer of protection. 343 00:20:58,427 --> 00:21:04,217 Howard: I've seen ransomware , you know, we think of ransomware as being on the 344 00:21:04,217 --> 00:21:05,957 order of the viruses we've dealt with. 345 00:21:07,342 --> 00:21:13,342 And the ransomware reports I see are much more frequently and this ransomware 346 00:21:13,342 --> 00:21:19,882 opened a door and then someone physically hacked for a long period of time. 347 00:21:20,422 --> 00:21:24,682 And they took over some workstation, eventually that some 348 00:21:24,682 --> 00:21:29,062 administrator logged into and they have an administrator password. 349 00:21:29,962 --> 00:21:35,212 And if we're just worried about, if we're just worried about the 350 00:21:35,212 --> 00:21:40,582 script kiddies in a, I can protect against the script kiddies in 351 00:21:40,582 --> 00:21:46,282 building my backup infrastructure and architecture and those permissions. 352 00:21:47,602 --> 00:21:51,442 But we're talking about more sophisticated attacks than that. 353 00:21:51,802 --> 00:21:55,942 And frankly we talk about it as ransomware, but it's also 354 00:21:55,942 --> 00:21:57,552 rogue administrator protection. 355 00:21:58,297 --> 00:22:03,637 Then it's also just the guy who is disgruntled and decides his 356 00:22:03,637 --> 00:22:06,487 way out the door, he's going to make life for his employer. 357 00:22:06,877 --> 00:22:08,287 You're protected against that too. 358 00:22:09,332 --> 00:22:09,632 curtis: Yeah. 359 00:22:09,662 --> 00:22:09,962 Yeah. 360 00:22:10,132 --> 00:22:13,832 And, sometimes rogue administrator is a true rogue administrator, meaning 361 00:22:13,832 --> 00:22:17,132 it's a, it's someone masquerading as an administrator as well. 362 00:22:18,272 --> 00:22:19,802 That hacker that you talked about. 363 00:22:20,432 --> 00:22:28,692 So let me let, me ask call it a difficult question, call it 364 00:22:28,692 --> 00:22:29,622 whatever you want to call it. 365 00:22:29,622 --> 00:22:38,292 But when I hear about boxes that where you're not supposed to be able to 366 00:22:38,292 --> 00:22:41,232 delete data, but then there is this other way where you can delete data. 367 00:22:42,222 --> 00:22:49,892 I immediately ask I, I have to ask the question doesn't that suggest 368 00:22:49,892 --> 00:22:56,622 that there is a this is, I'm assuming this is a, Unix-based OS and that 369 00:22:56,622 --> 00:22:58,422 there's that there is a root account, 370 00:22:58,542 --> 00:23:01,512 Howard: It we, run in containers under linux 371 00:23:01,861 --> 00:23:05,971 curtis: So there is an account, there is a a root account and that 372 00:23:06,001 --> 00:23:11,521 if someone did some sort of just the right attack against that box. 373 00:23:11,521 --> 00:23:15,441 And again you've already mentioned that there is that 374 00:23:15,441 --> 00:23:17,131 these are sophisticated attacks. 375 00:23:18,401 --> 00:23:23,456 If someone Did a privilege escalation attack against 376 00:23:23,996 --> 00:23:31,317 the CoreOS, and now they've gained access to a privileged Couldn't want 377 00:23:31,317 --> 00:23:32,067 Howard: if someone 378 00:23:32,216 --> 00:23:32,426 curtis: want. 379 00:23:34,998 --> 00:23:43,478 Howard: administrative access to the management network, because the 380 00:23:43,478 --> 00:23:45,938 ports that face users as storage 381 00:23:45,968 --> 00:23:47,588 ports, can't be logged into 382 00:23:51,856 --> 00:23:52,126 curtis: Okay. 383 00:23:52,126 --> 00:23:52,366 they're 384 00:23:52,366 --> 00:23:52,996 cause they're back. 385 00:23:52,996 --> 00:23:53,746 Cause they're backend, 386 00:23:56,283 --> 00:23:58,473 Howard: so if you're wondering, if you want to log into 387 00:23:58,473 --> 00:24:03,243 Linux as root on one of our appliances, then you need, 388 00:24:03,753 --> 00:24:07,353 then the management network has to be set, has to be compromised. 389 00:24:08,553 --> 00:24:14,553 And we start saying, are you looking for protection against destruction? 390 00:24:17,033 --> 00:24:23,723 Because if your data center is compromised, everything can be destroyed, 391 00:24:27,073 --> 00:24:30,973 but that's not really the level of attack that we're, concerned about. 392 00:24:35,133 --> 00:24:40,493 We're not talking about and someone walked into the data center because we 393 00:24:40,493 --> 00:24:44,123 hadn't disabled their key card and left 20 pounds of thermite in the middle of 394 00:24:44,123 --> 00:24:46,163 the floor, who would do such a thing. 395 00:24:49,078 --> 00:24:51,838 I've done that on video I was being paid. 396 00:24:54,358 --> 00:25:04,228 So you know, I, it is a vulnerability, but it's the 397 00:25:04,268 --> 00:25:06,238 generalest of the vulnerabilities. 398 00:25:06,238 --> 00:25:09,058 You're pointing out that if I have sufficient 399 00:25:09,448 --> 00:25:12,208 access, I can destroy anything. 400 00:25:14,851 --> 00:25:21,241 curtis: The but it sounds like you have protected from the rogue 401 00:25:21,241 --> 00:25:23,221 administrator, the stupid administrator. 402 00:25:25,441 --> 00:25:29,671 And and someone gaining access to those. 403 00:25:30,241 --> 00:25:34,591 But let me just you to clarify something from your previous answer, when you said 404 00:25:34,591 --> 00:25:38,281 that means the management network has been compromised, what do you mean by that? 405 00:25:40,608 --> 00:25:45,018 Howard: So you manage the system through different ethernet ports, 406 00:25:45,048 --> 00:25:47,288 then you access the system. 407 00:25:48,678 --> 00:25:54,128 And so too, you're if there's a vulnerability where a user could log 408 00:25:54,128 --> 00:26:02,768 into the appliance as the Linux root user that Linux root user can only 409 00:26:02,828 --> 00:26:06,578 log in on the management, physical Ethernet port on the appliance, not 410 00:26:06,578 --> 00:26:12,308 on the gigabit NVMe over fabric port. 411 00:26:13,361 --> 00:26:13,901 curtis: Gotcha. 412 00:26:13,991 --> 00:26:14,411 Okay. 413 00:26:14,558 --> 00:26:19,058 Howard: so network security should keep that from being an internet 414 00:26:19,058 --> 00:26:22,238 connected network and to attack. 415 00:26:23,001 --> 00:26:23,541 curtis: Gotcha. 416 00:26:23,601 --> 00:26:24,021 Gotcha. 417 00:26:24,591 --> 00:26:24,921 sense. 418 00:26:24,951 --> 00:26:25,371 Okay. 419 00:26:27,373 --> 00:26:27,613 Prasanna: I had a 420 00:26:27,613 --> 00:26:28,183 question. 421 00:26:29,533 --> 00:26:33,043 So Howard, before we dive more into the data protection side, one thing that 422 00:26:33,043 --> 00:26:38,743 was curious to me was you mentioned that vast supports file and object. 423 00:26:38,743 --> 00:26:41,233 Could you talk about some of the use cases that you see 424 00:26:41,233 --> 00:26:43,783 your customers using Vast Data? 425 00:26:43,783 --> 00:26:45,643 And then I think maybe some of the protection stuff will 426 00:26:45,643 --> 00:26:46,723 probably come alongside that. 427 00:26:47,223 --> 00:26:47,673 Howard: Sure. 428 00:26:49,083 --> 00:26:52,983 We have the majority of our customers use us for primary storage. 429 00:26:54,663 --> 00:26:59,973 And that includes one of the biggest travel sites who uses us for their 430 00:26:59,973 --> 00:27:06,333 big data analytics and are using the S3 Presto connectors to store 431 00:27:06,333 --> 00:27:11,613 all of their analytic data on us. 432 00:27:11,643 --> 00:27:15,933 So that we're much faster than a disk based object store, obviously. 433 00:27:15,933 --> 00:27:18,603 And they can do that processing faster. 434 00:27:19,263 --> 00:27:24,783 We have a lot of hedge funds who do time series analysis of trade 435 00:27:24,783 --> 00:27:28,683 data against large databases to try and predict the market. 436 00:27:29,553 --> 00:27:33,183 We have a lot of life sciences customers who are doing things like. 437 00:27:34,748 --> 00:27:42,218 Molecular modeling and cryo electron microscopy where one microscope generates 438 00:27:42,218 --> 00:27:46,808 many terabytes of data a day because we have very high resolution images. 439 00:27:48,638 --> 00:27:55,008 And we have a major motion picture studio who makes movies. 440 00:27:56,828 --> 00:28:00,578 Prasanna: And so it looks like they are using both sort of the file and the object 441 00:28:00,578 --> 00:28:03,158 interfaces for a lot of these use cases. 442 00:28:03,878 --> 00:28:07,868 So specifically around data protection and backup. 443 00:28:08,558 --> 00:28:12,973 A lot of times you hear The vendor's customers say, object 444 00:28:12,973 --> 00:28:14,203 store doesn't need to be backed up. 445 00:28:19,258 --> 00:28:29,498 Howard: This is a subject that personally I find myself on the fence about part 446 00:28:29,498 --> 00:28:36,803 of me goes I've built a huge amount of resiliency into this single system. 447 00:28:37,523 --> 00:28:44,183 And for durability, if for, availability, I may need to have it in another 448 00:28:44,183 --> 00:28:50,393 location, but for durability, assuming that the whole data center doesn't end 449 00:28:50,393 --> 00:28:56,033 up being a smoking hole in the ground I could get away without backing this up. 450 00:28:57,143 --> 00:29:01,433 I am N I remain firmly on the fence there. 451 00:29:03,743 --> 00:29:04,763 But 452 00:29:05,336 --> 00:29:10,526 curtis: assuming you have the second copy somewhere, you're going to 453 00:29:11,666 --> 00:29:12,116 write. 454 00:29:12,613 --> 00:29:16,153 Howard: may decide that it's, it is data that If, the whole data 455 00:29:16,153 --> 00:29:17,383 center goes away, I don't need. 456 00:29:17,956 --> 00:29:18,376 curtis: Okay. 457 00:29:18,586 --> 00:29:18,886 Yeah. 458 00:29:18,916 --> 00:29:19,306 Agreed. 459 00:29:19,516 --> 00:29:23,866 If, yeah, if we have That, data I would argue why did we make 460 00:29:23,866 --> 00:29:24,916 it in the first place, but, 461 00:29:24,933 --> 00:29:29,523 Howard: That the risk of that is the risk of that is small enough that I'm 462 00:29:29,523 --> 00:29:33,033 going to go once every thousand years this is going to cost me a million 463 00:29:33,033 --> 00:29:36,573 dollars, but it's going to cost me a million dollars a year to protect. 464 00:29:36,573 --> 00:29:37,803 So I'm going to take that risk. 465 00:29:38,796 --> 00:29:39,186 curtis: okay. 466 00:29:41,546 --> 00:29:45,246 So such I will agree to such data classes exist. 467 00:29:45,456 --> 00:29:47,376 I don't run into them much, but I will agree 468 00:29:47,453 --> 00:29:47,693 Howard: yeah. 469 00:29:47,753 --> 00:29:53,243 And and then we get to the okay, so this is the object store that does a 470 00:29:53,693 --> 00:29:59,633 deep dispersal coding, and they have three locations and I can lose one. 471 00:29:59,873 --> 00:30:01,403 So do I need to back that up? 472 00:30:02,603 --> 00:30:07,943 That starts getting really close to now I need to back it up because there could be 473 00:30:07,943 --> 00:30:09,743 a bug in the software that loses my data. 474 00:30:11,723 --> 00:30:14,363 'cause, that's the only thing that could cause that it's like 475 00:30:14,363 --> 00:30:17,373 unprotected against one of my three data centers being a smoking hole. 476 00:30:18,273 --> 00:30:26,088 what again, it's I could see you going, I want to be safe and I can 477 00:30:26,088 --> 00:30:27,338 see you going, it's not worth it. 478 00:30:28,061 --> 00:30:28,391 curtis: And. 479 00:30:29,408 --> 00:30:33,518 Howard: Now for us, most of our users use us for primary storage. 480 00:30:34,418 --> 00:30:40,058 And for someone like that, big data analytics data, they may not back it 481 00:30:40,058 --> 00:30:45,278 up because it's regenerate Hubble, and it's not actually in the form 482 00:30:45,278 --> 00:30:50,348 it's in on the object store, but it's extracts from other things and they 483 00:30:50,348 --> 00:30:55,058 can run the ETL again and it would be really annoying, but it is replaceable. 484 00:30:57,188 --> 00:31:01,088 And then we and then for other use cases this is primary data. 485 00:31:01,088 --> 00:31:02,078 I gotta protect it. 486 00:31:03,028 --> 00:31:07,838 And so we can do snapshots to an S3 compatible object store 487 00:31:08,348 --> 00:31:10,658 and back ourselves up that way. 488 00:31:11,888 --> 00:31:15,068 Or you can back us up the usual ways. 489 00:31:16,961 --> 00:31:24,371 curtis: And could you use one of the, like ones that are like 490 00:31:24,371 --> 00:31:27,761 glacier deep archive where I hope I don't ever have to use this. 491 00:31:27,761 --> 00:31:30,881 I know it's going to cost me a crap ton of money, but it'll save me a lot of money. 492 00:31:30,881 --> 00:31:33,191 In the meantime, can you use that kind of storage? 493 00:31:36,673 --> 00:31:40,873 Howard: The risk reading data out of that kind of storage 494 00:31:41,353 --> 00:31:43,723 requires a few manual steps. 495 00:31:44,983 --> 00:31:52,713 If you just use S3 standard then data in those snapshots is available 496 00:31:52,713 --> 00:31:57,013 in a .Remote folder, like the .Snapshots folder in the file system. 497 00:31:57,373 --> 00:32:01,993 So users can do self-service restore, but that required, but 498 00:32:02,173 --> 00:32:06,223 this, that feature means the object has to be immediately readable. 499 00:32:09,553 --> 00:32:11,713 And so if you, if it went to 500 00:32:11,713 --> 00:32:13,783 Glacier, then. 501 00:32:15,598 --> 00:32:19,558 And it would be like your net backup 502 00:32:19,643 --> 00:32:19,883 Prasanna: Okay. 503 00:32:20,158 --> 00:32:22,258 Howard: this backup isn't in the catalog anymore. 504 00:32:22,558 --> 00:32:26,758 So I got to put those files someplace where I can catalog it and then I got 505 00:32:26,758 --> 00:32:28,678 a catalog and then I can restore it. 506 00:32:30,608 --> 00:32:31,058 So if you 507 00:32:31,061 --> 00:32:32,501 curtis: so it's possible. 508 00:32:32,831 --> 00:32:32,891 it 509 00:32:32,891 --> 00:32:37,691 doesn't sound like it's very it's the smoking hole copy, right? 510 00:32:39,678 --> 00:32:41,158 Howard: It is annoying. 511 00:32:41,188 --> 00:32:44,338 But if it's just, but if you're protecting against the smoking hole, 512 00:32:44,878 --> 00:32:47,578 then you know, you may be willing to put up with the annoyance. 513 00:32:48,991 --> 00:32:51,661 curtis: I'm pretty sure we've said smoking hole more times 514 00:32:51,661 --> 00:32:53,941 than we've on this podcast. 515 00:32:53,971 --> 00:32:54,781 Just for the record. 516 00:32:55,351 --> 00:32:55,951 Just saying 517 00:32:57,331 --> 00:32:58,621 It's getting a lot of play today. 518 00:33:00,288 --> 00:33:02,828 Howard: I spent way too long as a disaster recovery planner. 519 00:33:03,891 --> 00:33:04,341 curtis: Yeah. 520 00:33:04,521 --> 00:33:05,031 Yeah. 521 00:33:08,631 --> 00:33:13,581 So the majority of your customers use you for primary storage, but clearly 522 00:33:13,581 --> 00:33:15,891 you're trying to expand your TAM, 523 00:33:15,953 --> 00:33:21,533 Howard: Well, w we, we deliver all flash at a substantially lower 524 00:33:21,533 --> 00:33:22,883 price than anybody else does. 525 00:33:24,143 --> 00:33:27,533 We start with using the cheapest QLC flash. 526 00:33:28,493 --> 00:33:34,403 We have a file system designed to treat that flash properly. 527 00:33:34,943 --> 00:33:40,853 So we never do small writes that would consume a lot of write amplification. 528 00:33:41,723 --> 00:33:45,353 We do very wide erasure code stripes. 529 00:33:45,443 --> 00:33:52,583 So we've got under 3% overhead, and then we do guaranteed better data reduction 530 00:33:52,583 --> 00:33:54,353 than anybody else in the business. 531 00:33:56,983 --> 00:34:05,983 And so that combination means that on an effective byte basis, from whatever backup 532 00:34:06,523 --> 00:34:10,733 data mover you're planning on using, we're going to be cheaper than a Data Domain. 533 00:34:11,658 --> 00:34:16,668 When you start saying that it's, you have more than a petabyte of data 534 00:34:16,698 --> 00:34:18,918 and you need multiple Data Domains. 535 00:34:19,308 --> 00:34:22,488 And each one of those is going to be a separate deduplication realm. 536 00:34:23,808 --> 00:34:26,628 Then the gap starts to grow substantially. 537 00:34:26,628 --> 00:34:30,948 So if so for these very large customers who have five or 10 or 20 538 00:34:30,948 --> 00:34:36,588 petabytes data across a bunch of Data Domains, simply the fact that we're 539 00:34:36,588 --> 00:34:43,608 one reduction realm makes that makes us much more efficient that can be. 540 00:34:44,208 --> 00:34:45,558 it's one system to manage. 541 00:34:45,558 --> 00:34:49,428 It's one namespace it's one 20 petabytes or 50 petabytes system. 542 00:34:53,906 --> 00:34:56,066 curtis: So you're saying, so let me just make sure I understood 543 00:34:56,066 --> 00:34:57,506 what you said there correctly. 544 00:34:59,066 --> 00:35:03,326 saying on a, regardless of the size of the system, you should 545 00:35:03,326 --> 00:35:06,946 be priced competitive with a Data Domain, but then the bigger you get, 546 00:35:07,226 --> 00:35:07,946 better you look. 547 00:35:08,478 --> 00:35:13,758 Howard: under about 500, any pricing experiments under about 500 terabytes, 548 00:35:14,006 --> 00:35:14,426 curtis: Okay. 549 00:35:14,576 --> 00:35:14,996 Okay. 550 00:35:15,108 --> 00:35:16,828 Howard: in the large end of the business, but yes. 551 00:35:16,828 --> 00:35:20,876 curtis: Right, That is interesting though, that sort of. 552 00:35:22,381 --> 00:35:24,691 into that end of the business. 553 00:35:25,261 --> 00:35:32,941 And you had another there was another large, all flash competitor that's 554 00:35:32,941 --> 00:35:37,681 doing very well, but they have a very different architecture, they're referring 555 00:35:37,681 --> 00:35:39,331 of course, to the orange company. 556 00:35:40,651 --> 00:35:40,981 And 557 00:35:41,898 --> 00:35:42,738 Howard: Yeah, but there, 558 00:35:42,931 --> 00:35:43,321 curtis: than you. 559 00:35:45,258 --> 00:35:50,508 Howard: If you're talking about Flash Blade, that's really a shared nothing 560 00:35:50,508 --> 00:35:57,588 architecture it's of being pizza box servers, they're blade servers, and each 561 00:35:57,588 --> 00:36:07,628 blade has flash modules built in And they they don't scale nearly as large. 562 00:36:09,687 --> 00:36:13,287 curtis: So it sounds like you, you just took, you've built an 563 00:36:13,287 --> 00:36:17,397 architecture based on several new pieces of technology that simply 564 00:36:17,397 --> 00:36:20,977 weren't available, say, five years ago, 565 00:36:21,264 --> 00:36:21,624 Howard: Yeah. 566 00:36:23,394 --> 00:36:33,234 We, are the storage system designed from a clean slate around the 2016 toolbox. 567 00:36:33,834 --> 00:36:35,604 So QLC, flash, 568 00:36:36,474 --> 00:36:46,164 SCM, NVMe over fabrics and other people shoe horn one or two of those technologies 569 00:36:46,164 --> 00:36:51,834 into an existing architecture, but we built the whole architecture 570 00:36:51,924 --> 00:36:54,654 around having those technologies. 571 00:36:56,134 --> 00:37:02,464 Yeah, putting all of the metadata in SCM with no cache meant it had to be in SCM. 572 00:37:02,914 --> 00:37:07,744 And it meant the connection between the compute server and that SCM had to be 573 00:37:07,744 --> 00:37:13,624 fast enough that we weren't going if we cached this, it would be a lot faster. 574 00:37:13,954 --> 00:37:16,894 So that meant it had to be NVMe over fabrics. 575 00:37:17,674 --> 00:37:20,284 And then the QLC flash gives us the cost. 576 00:37:20,631 --> 00:37:28,911 But it, really is if you look at any storage system, it's by definition built 577 00:37:28,941 --> 00:37:34,161 with the parts that the industry is making when they sat down to design it. 578 00:37:35,764 --> 00:37:36,154 curtis: Yeah. 579 00:37:37,701 --> 00:37:46,851 Howard: And that when x86 processor when Mahalum came along and the 580 00:37:46,851 --> 00:37:53,331 memory bandwidth and the number of PCI e-lanes on processors got big enough. 581 00:37:53,331 --> 00:37:57,831 All of a sudden we stopped seeing FPGAs and ASICs in storage systems, we started 582 00:37:57,831 --> 00:38:02,031 seeing software defined storage, cause what was available for the designers 583 00:38:02,031 --> 00:38:10,251 changed and the NVMe over fabrics has been used by most of the storage 584 00:38:10,251 --> 00:38:14,121 vendors for that last mile connection going well, it's going to be fast and 585 00:38:14,121 --> 00:38:20,181 then fiber channel or iSCSI for the user machine to access the storage. 586 00:38:20,961 --> 00:38:26,631 But it hasn't been as effectively used for the server that is the logical 587 00:38:26,631 --> 00:38:35,811 controller to access the media on the back end and the way we use it, we broke the 588 00:38:36,471 --> 00:38:42,141 traditional limitation that a drive had to be owned by one or two controllers. 589 00:38:42,621 --> 00:38:47,841 Cause I drive a SAS drive where an NVMe drive has one or two ports. 590 00:38:47,841 --> 00:38:48,101 Prasanna: Yea. 591 00:38:49,261 --> 00:38:56,551 Howard: We connect that NVMe SSD to what we call a fabric module, which 592 00:38:56,551 --> 00:38:59,401 is an NVMe over fabrics router. 593 00:38:59,401 --> 00:39:06,091 And in fact, in the new box, it's going to be a pair of Nvidia Bluefield cards 594 00:39:07,741 --> 00:39:12,961 and the Bluefield card routes, NVMe over fabrics requests from the ethernet network 595 00:39:13,321 --> 00:39:16,891 to the SSDs and routes the responses back. 596 00:39:16,921 --> 00:39:18,301 But that's all it does. 597 00:39:18,331 --> 00:39:22,321 We don't need x86 servers in the enclosure. 598 00:39:22,321 --> 00:39:25,101 We can do it on the ARMs and the offloads and the Bluefields. 599 00:39:25,126 --> 00:39:27,226 Prasanna: and these are the DPUs, correct? 600 00:39:27,946 --> 00:39:28,486 Howard: Yes. 601 00:39:28,516 --> 00:39:28,726 Yeah. 602 00:39:28,726 --> 00:39:36,346 The Bluefield is, the DPU it's the Nvidia Mellanox version of that. 603 00:39:36,436 --> 00:39:42,436 And so it has an ARM some ARM cores and NVMe over fabrics and RDMA and 604 00:39:42,436 --> 00:39:44,776 other built-in offloads in the chip. 605 00:39:45,586 --> 00:39:49,786 And so we leverage that to do the routing of requests from the front 606 00:39:49,786 --> 00:39:57,976 end servers, everything is, all the work gets done the SSDs and get that 607 00:39:58,216 --> 00:40:01,276 clean fast, more cost-effective channel 608 00:40:03,954 --> 00:40:09,624 curtis: Let me go back in time when you did that first presentation that 609 00:40:09,624 --> 00:40:11,874 you did to the Storage Field Day folks, 610 00:40:12,566 --> 00:40:12,716 Howard: Yep. 611 00:40:13,854 --> 00:40:16,904 curtis: how did that go over with, with those folks? 612 00:40:16,904 --> 00:40:18,061 Howard: It went over pretty well. 613 00:40:18,181 --> 00:40:23,431 There was a little being from Missouri and, 614 00:40:23,461 --> 00:40:23,581 you, 615 00:40:23,581 --> 00:40:24,391 know, we should show you, 616 00:40:24,674 --> 00:40:25,874 curtis: Cause you weren't because you were brand new. 617 00:40:25,874 --> 00:40:26,384 at that point. 618 00:40:26,804 --> 00:40:28,231 Howard: We We were brand new. 619 00:40:29,581 --> 00:40:34,471 And now we're going, okay, look, we've sold a couple of exabytes of storage. 620 00:40:36,341 --> 00:40:42,281 Now at this we, our go to market model's a little different, we sell software. 621 00:40:42,611 --> 00:40:49,241 We arrange for customers to buy the pre-approved hardware at cost. 622 00:40:50,801 --> 00:40:51,251 And the 623 00:40:51,251 --> 00:40:53,021 software licenses are, 624 00:40:53,099 --> 00:40:53,939 curtis: a little interesting. 625 00:40:54,671 --> 00:40:56,831 Howard: and the software licenses are transferable. 626 00:40:58,301 --> 00:41:02,231 So you license a petabyte of software. 627 00:41:03,091 --> 00:41:06,901 And you upgrade the hardware when you feel like you're want to upgrade the hardware. 628 00:41:06,901 --> 00:41:11,461 Cause you want the denser faster one that is always coming, but we'll write 629 00:41:11,461 --> 00:41:16,041 the support contract for 10 years for any appliance from install date. 630 00:41:17,981 --> 00:41:18,331 So 631 00:41:18,331 --> 00:41:19,501 Prasanna: That's very different 632 00:41:21,001 --> 00:41:22,021 Howard: well, a typical 633 00:41:22,021 --> 00:41:27,061 vendor, you would buy an appliance, it would come with an oEM software license. 634 00:41:27,721 --> 00:41:29,611 They would write five years of support. 635 00:41:30,901 --> 00:41:35,011 And in year six they would encourage you very strongly to rebuy. 636 00:41:35,401 --> 00:41:35,701 Prasanna: yep. 637 00:41:37,081 --> 00:41:41,251 Howard: And then when you rebuy, you have to buy another appliance the 638 00:41:41,251 --> 00:41:43,021 software license isn't transferable. 639 00:41:43,831 --> 00:41:45,481 So you have to buy another software license. 640 00:41:46,801 --> 00:41:53,251 So with us, you gotta have your VAR go to a VAR, a hundred 641 00:41:53,251 --> 00:41:54,671 percent channel you go to a VAR. 642 00:41:55,561 --> 00:42:01,291 your VAR, goes to Avnet, says, I want this hardware for Vast. 643 00:42:02,831 --> 00:42:06,611 Now $1.2 million average selling price. 644 00:42:07,151 --> 00:42:08,981 One of our sales guys is involved. 645 00:42:09,011 --> 00:42:12,281 We're writing the high touch sale. 646 00:42:12,281 --> 00:42:15,911 It's not somebody went on a website someplace. 647 00:42:17,111 --> 00:42:27,521 Um, but essentially the VAR, writes two POs: one to Avnet for the hardware and one 648 00:42:27,521 --> 00:42:34,751 to us for the actually he writes one PO to Avnet, Avnet cuts us a PO for the software 649 00:42:37,001 --> 00:42:41,131 and, that's a capacity subscription. 650 00:42:43,171 --> 00:42:51,101 So if you bought a 675 terabyte, enclosure and an appliance, that's got 651 00:42:51,101 --> 00:42:55,871 four servers that provide the front end, which is our usual entry point. 652 00:42:56,801 --> 00:42:59,181 You could license a hundred terabytes for a year. 653 00:43:00,861 --> 00:43:03,561 Multiples of a hundred terabytes for multiples a year. 654 00:43:06,714 --> 00:43:10,734 curtis: And so that, I think that addresses the question that I had. 655 00:43:10,734 --> 00:43:15,864 Cause I listened to the Chris Evans podcasts that you guys did. 656 00:43:16,901 --> 00:43:17,291 Howard: Yeah. 657 00:43:17,334 --> 00:43:25,674 curtis: and there was this talk of the 10 year And, again I'm gonna, I'm gonna just 658 00:43:25,961 --> 00:43:26,201 Howard: Perfect. 659 00:43:26,514 --> 00:43:31,194 curtis: acknowledge that I live in a SaaS world where we preach against 660 00:43:31,974 --> 00:43:36,594 large capacity licensing and capital purchases and all of that stuff. 661 00:43:37,014 --> 00:43:40,829 So when I heard 10 year purchase. 662 00:43:41,069 --> 00:43:41,969 I was like, what? 663 00:43:41,969 --> 00:43:46,499 I gotta, I got to decide now how much I need for 10 years, but that doesn't 664 00:43:46,499 --> 00:43:47,679 sound like what you're talking about. 665 00:43:47,679 --> 00:43:48,219 Howard: No, No, no. 666 00:43:48,271 --> 00:43:48,541 no. 667 00:43:49,471 --> 00:43:51,421 You th you buy the hardware. 668 00:43:52,589 --> 00:43:52,919 curtis: Right. 669 00:43:52,919 --> 00:43:58,071 Howard: We will write a support contract and software license. 670 00:43:58,071 --> 00:43:58,961 One agreement. 671 00:43:59,461 --> 00:44:05,561 For that hardware for up to 10 years from install date at the same rate. 672 00:44:08,971 --> 00:44:12,211 So if you want to keep it for 10 years, you keep it for 10 years 673 00:44:14,499 --> 00:44:14,829 Bought 674 00:44:15,219 --> 00:44:18,039 curtis: I could buy a smaller one and then add capacity. 675 00:44:19,341 --> 00:44:19,941 Howard: Oh yeah. 676 00:44:22,181 --> 00:44:23,961 Our NRR is three. 677 00:44:27,401 --> 00:44:30,281 Lots of people buy small and add capacity. 678 00:44:35,661 --> 00:44:37,281 We had a 300% NRR. 679 00:44:37,841 --> 00:44:38,601 Prasanna: I think you meant NRR, 680 00:44:38,771 --> 00:44:39,131 right? 681 00:44:39,274 --> 00:44:40,294 curtis: Thanks for explaining. 682 00:44:42,074 --> 00:44:42,434 Yeah. 683 00:44:42,494 --> 00:44:43,274 NRR, 684 00:44:43,934 --> 00:44:44,954 you said ARR. 685 00:44:45,224 --> 00:44:45,674 That's why you 686 00:44:45,674 --> 00:44:46,814 have me confused there for a minute. 687 00:44:47,471 --> 00:44:47,771 Howard: Yeah 688 00:44:48,674 --> 00:44:51,044 curtis: I was like an annual recurring revenue of three, three. 689 00:44:52,694 --> 00:44:54,924 Met meant net retention rate, you're saying? 690 00:44:55,674 --> 00:44:56,094 yeah. 691 00:44:56,094 --> 00:45:01,794 So you're saying 300% your customers start out at X and they end up 692 00:45:01,794 --> 00:45:04,884 with three X very regularly. 693 00:45:05,874 --> 00:45:06,264 Okay. 694 00:45:08,464 --> 00:45:09,596 Howard: You know, and you can do that. 695 00:45:09,924 --> 00:45:11,754 curtis: it just grows as they need it to grow. 696 00:45:12,416 --> 00:45:12,716 Howard: Yeah. 697 00:45:12,746 --> 00:45:17,606 And you can do it in the hardware, so if you want to start really small, then 698 00:45:17,606 --> 00:45:21,156 you can buy hardware and license it 699 00:45:21,651 --> 00:45:22,521 Prasanna: oh, interesting. 700 00:45:22,886 --> 00:45:28,376 Howard: So You can buy, a 600 terabyte box and a hundred terabytes software 701 00:45:28,376 --> 00:45:34,586 license, and the 600 terabyte box you bought at what would be our cost. 702 00:45:34,586 --> 00:45:38,996 If we were still selling hardware, we negotiate the cost with the intel 703 00:45:38,996 --> 00:45:42,026 and key Aksia and those vendors. 704 00:45:42,121 --> 00:45:43,681 Prasanna: so you used to sell hardware and then you 705 00:45:43,921 --> 00:45:44,161 of, 706 00:45:44,336 --> 00:45:45,956 Howard: started off in an appliance model. 707 00:45:49,554 --> 00:45:50,594 curtis: Why would I do that? 708 00:45:51,164 --> 00:45:54,584 Is that just like ease of large capital purchase thing? 709 00:45:55,301 --> 00:45:55,601 Howard: Yeah. 710 00:45:56,324 --> 00:45:56,534 curtis: why 711 00:45:56,534 --> 00:45:57,674 would I buy a bigger box 712 00:45:57,701 --> 00:46:03,131 Howard: university, we had a university had this much money in this year's budget. 713 00:46:04,164 --> 00:46:04,854 curtis: Oh, okay. 714 00:46:05,381 --> 00:46:09,221 Howard: We won't put more than a hundred terabytes on it before the next budget 715 00:46:09,221 --> 00:46:13,811 comes around when we renew, we'll renew it as a 400 terabyte license. 716 00:46:15,671 --> 00:46:18,041 Prasanna: and I think this is where at the beginning, you said Howard, that you're 717 00:46:18,041 --> 00:46:21,791 looking at releasing a smaller unit. 718 00:46:22,311 --> 00:46:22,521 Howard: Yeah. 719 00:46:22,551 --> 00:46:24,231 So the new box is one. 720 00:46:24,231 --> 00:46:24,621 You, 721 00:46:25,281 --> 00:46:30,531 it uses the ESS F one L the ruler form factor as, DS. 722 00:46:31,101 --> 00:46:38,661 So we can, we have 2215 terabyte SSDs for 3 38 raw bat, 300 usable. 723 00:46:40,701 --> 00:46:47,301 And that's half the physical size, half the capacity, because what we 724 00:46:47,301 --> 00:46:51,321 have now, it holds 56 SSDs and two U 725 00:46:51,801 --> 00:46:52,251 Prasanna: Gotcha. 726 00:46:54,271 --> 00:47:00,211 Howard: Yeah, the new one is, from the fabric module is those NVMe routers today. 727 00:47:00,931 --> 00:47:02,761 Each one has to be a dual Xeon. 728 00:47:02,761 --> 00:47:04,021 So we have enough PCIE 729 00:47:04,021 --> 00:47:08,941 lanes and the processors don't do hardly anything. 730 00:47:09,391 --> 00:47:11,671 So there's just there's costs there. 731 00:47:11,671 --> 00:47:13,621 We don't need, if the Bluefield 732 00:47:13,621 --> 00:47:13,951 thing 733 00:47:15,271 --> 00:47:15,811 Prasanna: That's exciting. 734 00:47:15,979 --> 00:47:16,339 curtis: right. 735 00:47:18,349 --> 00:47:21,769 So let's, focus for a little bit on. 736 00:47:24,984 --> 00:47:29,394 The only reason I have historically been when, I historically heard the 737 00:47:29,394 --> 00:47:35,424 idea of using flash for backup, I'm like, that sounds ridiculous because 738 00:47:35,754 --> 00:47:44,694 for the same for cost reasons, too expensive I'm hearing you that so I 739 00:47:44,694 --> 00:47:49,194 would put it this way that, in, in this upcoming world, in this current world 740 00:47:51,174 --> 00:47:57,114 in a world where we have large nation states invading other nation states 741 00:47:57,384 --> 00:48:04,674 and then large ransomware organizations in those countries, we had this, was 742 00:48:04,674 --> 00:48:08,874 our last th they're talking about. 743 00:48:08,934 --> 00:48:13,704 So we're, talking about being retaliated against because of this other country. 744 00:48:13,824 --> 00:48:14,514 It's crazy. 745 00:48:15,084 --> 00:48:25,084 So you have this this, need more than ever before for large recoveries. 746 00:48:25,174 --> 00:48:28,654 And I, do believe strongly that there's really only one of two 747 00:48:28,654 --> 00:48:33,934 ways to be really successful in any sort of ransomware situation. 748 00:48:35,484 --> 00:48:38,524 And, it's basically about fighting the laws of physics .Either you 749 00:48:38,524 --> 00:48:40,174 have to have already restored it. 750 00:48:40,564 --> 00:48:48,364 So you already have a hot standby ready to go to switch over to or you're 751 00:48:48,364 --> 00:48:54,214 doing live mount directly from your backup and live mount directly from 752 00:48:54,214 --> 00:49:00,424 your backup is only going to happen if you either aren't, deduplicating 753 00:49:02,014 --> 00:49:04,354 like, the way Data Domain does, or 754 00:49:04,461 --> 00:49:04,561 Howard: Right. 755 00:49:04,761 --> 00:49:05,944 curtis: have flash as far 756 00:49:06,334 --> 00:49:06,664 Tell. 757 00:49:06,961 --> 00:49:11,581 Howard: if you're not, even if you're not, deduplicating when you start talking 758 00:49:11,581 --> 00:49:15,391 about big, hard drives the IO density just 759 00:49:15,391 --> 00:49:18,121 isn't there it's better 760 00:49:19,129 --> 00:49:22,779 curtis: Some somewhere between you and Data Domain, I would put Exagrid, 761 00:49:23,449 --> 00:49:25,189 because exa grid has that front end. 762 00:49:25,189 --> 00:49:27,979 It's not de duplicated now they're there. 763 00:49:28,189 --> 00:49:30,019 They're nowhere near the size of you. 764 00:49:30,271 --> 00:49:30,781 Howard: right, no. 765 00:49:30,821 --> 00:49:35,001 And they have some, and they, have, some flash cache. 766 00:49:35,021 --> 00:49:39,281 And if you look at guys who do integrated appliances where the 767 00:49:39,281 --> 00:49:43,601 software and the target are one thing, those are typically hybrids. 768 00:49:44,751 --> 00:49:48,941 And, so they'll do an instant recover for one or two VMs pretty well. 769 00:49:48,941 --> 00:49:50,141 Cause there's enough flash for that. 770 00:49:51,331 --> 00:49:57,301 But when you start going, I need the database server behind my ERP, instant 771 00:49:57,301 --> 00:50:05,131 recovered, or I need all 50 of these VMs, instant recovered, then it's then you 772 00:50:05,131 --> 00:50:08,041 just, don't have enough flash and you're going to get hard drive performance, 773 00:50:09,599 --> 00:50:09,989 curtis: And so 774 00:50:10,019 --> 00:50:13,229 what it sounds like you've replaced the hard drives with QLC 775 00:50:14,221 --> 00:50:14,581 Howard: right, 776 00:50:15,499 --> 00:50:18,189 curtis: Help me because I don't live in this world QLC from 777 00:50:18,189 --> 00:50:21,939 a cost perspective regular. 778 00:50:22,231 --> 00:50:24,441 Howard: it's, not just QLC. 779 00:50:24,741 --> 00:50:29,091 So QLC means quad level cell holds four bits per cell. 780 00:50:30,319 --> 00:50:30,739 curtis: okay? 781 00:50:31,131 --> 00:50:35,061 Howard: The more, bits you hold, the closer, the voltage levels 782 00:50:35,061 --> 00:50:40,011 that represent the differences are, and the more sensitive the cells 783 00:50:40,011 --> 00:50:43,341 become to a few electrons escaping. 784 00:50:45,221 --> 00:50:48,761 If you have SLC, it's like a light switch it's on or off, 785 00:50:51,101 --> 00:50:53,411 It doesn't matter if a few electrons escape, you can still 786 00:50:53,411 --> 00:50:54,761 tell whether it's on or off. 787 00:50:55,931 --> 00:50:56,561 QLC. 788 00:50:56,561 --> 00:50:58,001 You got 16 values. 789 00:50:59,791 --> 00:51:04,861 The difference between value 13 and value 14 might only be a handful of electrons. 790 00:51:06,661 --> 00:51:09,961 So QLC has less endurance. 791 00:51:09,961 --> 00:51:13,171 Cause every time you erase it, the insulating layers wear down 792 00:51:13,171 --> 00:51:16,441 a little and a few more electrons have opportunities to escape. 793 00:51:18,241 --> 00:51:23,251 And it's slower to write because you have to adjust the voltage level just right 794 00:51:23,251 --> 00:51:25,021 to be one of those 16 voltage levels. 795 00:51:25,261 --> 00:51:26,761 And that takes a little bit longer. 796 00:51:28,111 --> 00:51:30,871 Now the slower to write, we don't really care about because 797 00:51:31,531 --> 00:51:34,351 we acknowledge the writes while it's still in the SCM. 798 00:51:35,551 --> 00:51:41,041 So as long as we are flushing that data out of the SCM, in bandwidth terms 799 00:51:41,041 --> 00:51:44,411 fast enough, Latency is unimportant. 800 00:51:46,271 --> 00:51:50,081 and the endurance we specifically do a lot of things in our 801 00:51:50,081 --> 00:51:52,691 software to manage endurance. 802 00:51:53,831 --> 00:52:01,391 So we write very large writes so that the SSD doesn't have to garbage collect 803 00:52:01,391 --> 00:52:04,791 internally to accommodate small writes. 804 00:52:05,121 --> 00:52:12,381 We erase very large erases so that we delete all of the data in an erase block 805 00:52:12,381 --> 00:52:16,701 in the flash so that the SSD doesn't have to garbage collect internally. 806 00:52:17,631 --> 00:52:22,731 And that means not only can we use QLC, but we can use dirt cheap QLC 807 00:52:23,001 --> 00:52:31,531 SSDs that don't have a DRAM buffer in them to protect the QLC from wear. 808 00:52:33,486 --> 00:52:37,686 If you have a DRAM buffer, then you can aggregate multiple small 809 00:52:37,686 --> 00:52:42,696 writes, but yet, but now if power fails, it's DRAM, you lose the data. 810 00:52:42,996 --> 00:52:47,196 So you need a power fail protection circuit, and you need big capacitors 811 00:52:47,406 --> 00:52:49,116 to power, the power fail protection 812 00:52:49,116 --> 00:52:54,366 circuit so that you can that you can dump the DRAM into flash and 813 00:52:54,396 --> 00:52:55,986 right, and it all starts to add up. 814 00:52:56,646 --> 00:53:01,416 So the SSDs we buy, the other customers are hyperscalers. 815 00:53:02,166 --> 00:53:03,636 They put them in servers. 816 00:53:04,416 --> 00:53:08,256 They only need one port they're writing long tail data. 817 00:53:08,256 --> 00:53:10,476 It's not like they're overriding this stuff all the time. 818 00:53:10,776 --> 00:53:14,196 It's just too many people are looking at that drunken fat frat 819 00:53:14,196 --> 00:53:18,666 boy picture on Facebook it to be on disk so it's on flash. 820 00:53:20,379 --> 00:53:21,699 curtis: A. 821 00:53:23,356 --> 00:53:29,016 Howard: We're leveraging all of that to keep so that we can literally 822 00:53:29,016 --> 00:53:30,826 use that lowest cost flash. 823 00:53:32,271 --> 00:53:37,161 And do the 10 year support because the 10 year support includes if the 824 00:53:37,161 --> 00:53:38,751 SSD wears out, we'll replace it. 825 00:53:41,566 --> 00:53:45,106 Prasanna: cause normally QLC isn't rated for that long. 826 00:53:45,166 --> 00:53:45,586 I believe. 827 00:53:45,586 --> 00:53:45,926 Right. 828 00:53:45,926 --> 00:53:47,886 SLC is years 829 00:53:48,046 --> 00:53:53,241 Howard: S SLC SLC is the very high endurance flesh, but the typical 830 00:53:53,241 --> 00:53:57,891 flash that you see for volume use today is TLC triple level cell. 831 00:53:58,341 --> 00:54:00,201 So it's three bits instead of four bits. 832 00:54:00,531 --> 00:54:04,911 So QLC is 30% cheaper to make because it holds more bits per cell. 833 00:54:07,521 --> 00:54:12,121 And QLC has substantially less endurance. 834 00:54:13,281 --> 00:54:16,481 So when you start looking at enterprise SSDs on newegg. 835 00:54:19,041 --> 00:54:26,321 The 0.1 drive write per day, SSD is slightly better than the ones we use. 836 00:54:27,951 --> 00:54:33,706 And the three drive write per day, SSD, you notice has less capacity because 837 00:54:33,706 --> 00:54:35,416 it's got the same amount of flash. 838 00:54:35,716 --> 00:54:39,886 It's just more over-provisioned so they can wear level across more of it. 839 00:54:40,336 --> 00:54:44,926 And the three drive rate per day, SSD probably has a DRAM cache 840 00:54:44,926 --> 00:54:46,156 and all this stuff to protect it. 841 00:54:46,231 --> 00:54:46,411 Prasanna: Yeah 842 00:54:47,806 --> 00:54:53,506 Howard: And that's what most enterprise storage systems need because how 843 00:54:53,506 --> 00:54:58,756 they put the data in the drive dates back to when it was a disk drive. 844 00:54:59,266 --> 00:55:05,146 And you were trying to keep data logically adjacent, not try and manage 845 00:55:05,176 --> 00:55:07,536 the write pool inside the drive. 846 00:55:07,536 --> 00:55:07,606 Prasanna: yeah, 847 00:55:07,696 --> 00:55:09,056 Howard: The requirements were different. 848 00:55:09,984 --> 00:55:10,224 curtis: Yeah. 849 00:55:10,224 --> 00:55:10,674 Interesting. 850 00:55:10,674 --> 00:55:10,794 Yeah. 851 00:55:10,824 --> 00:55:11,874 So again, going back to. 852 00:55:11,874 --> 00:55:17,424 the fact that you built this from the scratch with that toolbox 853 00:55:17,454 --> 00:55:21,384 from 2016, and you were like we need to, manage write leveling, 854 00:55:21,651 --> 00:55:26,931 Howard: And look, our founder Renen Hallak was the chief engineer at Extreme IO. 855 00:55:28,251 --> 00:55:33,171 And when he got tired of working for Michael Dell, he got to talk to Extreme IO 856 00:55:33,171 --> 00:55:34,731 customers and find out what they wanted. 857 00:55:35,931 --> 00:55:40,061 And nobody said we want faster, Extreme IO was already all flash. 858 00:55:40,061 --> 00:55:41,991 They were still adjusting to all flash. 859 00:55:43,641 --> 00:55:47,001 And it was plenty fast, but everybody wanted to be able to use 860 00:55:47,001 --> 00:55:48,621 that all flash for more things. 861 00:55:50,271 --> 00:55:57,111 And so our whole system is designed to provide very high, random read 862 00:55:57,351 --> 00:56:02,991 performance, across large amounts of flash at an affordable price. 863 00:56:04,884 --> 00:56:05,304 curtis: Got it. 864 00:56:06,841 --> 00:56:09,811 Howard: And so our our performance asymmetry is exactly 865 00:56:09,811 --> 00:56:11,371 the opposite of data domains. 866 00:56:14,694 --> 00:56:16,704 curtis: wait, explain what you just said. 867 00:56:17,451 --> 00:56:22,821 Howard: Our performance asymmetry is exactly the opposite of data domains. 868 00:56:23,751 --> 00:56:26,121 They don't publish restore speeds anymore. 869 00:56:26,151 --> 00:56:32,391 Haven't for years we publish, read speeds and writes speeds and reads 870 00:56:32,391 --> 00:56:34,281 are at eight times faster than rights. 871 00:56:34,801 --> 00:56:36,511 Prasanna: That doesn't mean your rights are slow either. 872 00:56:36,661 --> 00:56:37,051 Just for 873 00:56:38,031 --> 00:56:44,601 Howard: No Our, smallest system does five gigabytes per second of rights. 874 00:56:46,881 --> 00:56:47,091 Yeah. 875 00:56:47,121 --> 00:56:54,411 Or your story system probably doesn't keep up with that, but that's the SLOs. 876 00:56:54,531 --> 00:56:59,211 But what that means is if you scale a system the traditional way, and 877 00:56:59,211 --> 00:57:03,561 you say, I need to move this many terabytes over this many hours, so you 878 00:57:03,561 --> 00:57:05,511 have to scale it by right performance. 879 00:57:07,041 --> 00:57:10,041 Your backups are going to be much faster than your restores. 880 00:57:12,441 --> 00:57:12,801 Excuse me. 881 00:57:12,801 --> 00:57:13,761 your restores are much 882 00:57:13,761 --> 00:57:14,301 faster than your 883 00:57:14,301 --> 00:57:14,781 backups. 884 00:57:14,791 --> 00:57:15,061 Prasanna: Yeah, 885 00:57:15,951 --> 00:57:16,191 Howard: Yeah 886 00:57:16,581 --> 00:57:18,411 we read much faster than we write. 887 00:57:18,411 --> 00:57:21,711 And so if you size for backups speed, you're a store. 888 00:57:21,711 --> 00:57:22,611 Speed's going to be 889 00:57:22,794 --> 00:57:23,034 curtis: yeah. 890 00:57:23,391 --> 00:57:23,991 Howard: nice. 891 00:57:27,139 --> 00:57:27,529 curtis: All right. 892 00:57:29,179 --> 00:57:30,619 Consider me impressed, Howard. 893 00:57:31,699 --> 00:57:32,189 you know, I, 894 00:57:32,211 --> 00:57:32,961 Prasanna: do by the way 895 00:57:33,681 --> 00:57:33,711 curtis: I 896 00:57:34,201 --> 00:57:34,721 Howard: I I've 897 00:57:34,989 --> 00:57:35,559 curtis: I, I, 898 00:57:35,731 --> 00:57:36,181 Howard: time. 899 00:57:36,181 --> 00:57:37,621 I've impressed him once. 900 00:57:38,221 --> 00:57:39,571 this is makes twice. 901 00:57:39,571 --> 00:57:40,981 I'm really, I'm happy with that, 902 00:57:42,169 --> 00:57:47,749 curtis: yeah it sounds like you're, clearly you've been 903 00:57:47,749 --> 00:57:48,799 in the business a long time. 904 00:57:48,799 --> 00:57:51,589 You've seen those companies that have really interesting technology 905 00:57:51,769 --> 00:57:53,029 and nobody's buying anything. 906 00:57:53,539 --> 00:57:55,099 You're not that you, 907 00:57:55,291 --> 00:57:55,621 Howard: but 908 00:57:55,819 --> 00:57:59,239 curtis: the really interesting technology, but you're also actually selling it, 909 00:57:59,539 --> 00:57:59,929 right? 910 00:58:00,171 --> 00:58:01,791 Howard: I decided it was time to get a job. 911 00:58:03,261 --> 00:58:06,021 And I talked to the folks at Vast, who were still in stealth. 912 00:58:07,371 --> 00:58:09,711 And I said to myself, look, Howard, you're a storyteller. 913 00:58:10,911 --> 00:58:12,321 And this is a really good story. 914 00:58:14,141 --> 00:58:16,211 And it doesn't matter whether it succeeds or not. 915 00:58:17,231 --> 00:58:18,761 You're going to have a good story to tell. 916 00:58:21,451 --> 00:58:25,321 and low and behold, it's one of those cases where it was a good 917 00:58:25,321 --> 00:58:28,351 story and the market requirement fit. 918 00:58:30,991 --> 00:58:31,741 And 919 00:58:31,821 --> 00:58:33,504 curtis: don't have to create the need. 920 00:58:34,361 --> 00:58:41,261 Howard: we are selling we have, for the past couple of years 921 00:58:41,261 --> 00:58:44,681 done comparisons, all the storage companies have gone public you. 922 00:58:44,751 --> 00:58:44,921 Yeah. 923 00:58:44,931 --> 00:58:46,691 We're growing faster than all of them put together 924 00:58:48,611 --> 00:58:52,051 curtis: all right Howard thanks a lot for coming on. 925 00:58:52,091 --> 00:58:53,501 We might have to have you back. 926 00:58:53,531 --> 00:58:59,981 Cause I, I know that I know we've, just begun to scratch the surface and but 927 00:59:00,821 --> 00:59:02,471 sounds like you got a good gig over there. 928 00:59:02,621 --> 00:59:03,131 I'm glad. 929 00:59:03,311 --> 00:59:04,331 Both of us could be 930 00:59:04,991 --> 00:59:05,621 employed. 931 00:59:07,323 --> 00:59:07,743 Howard: Ed 932 00:59:07,871 --> 00:59:08,321 curtis: well. 933 00:59:08,703 --> 00:59:10,293 Howard: for the people have known us a long time. 934 00:59:10,293 --> 00:59:18,003 It really must be shocking to you and I both the same job multiple years, but 935 00:59:19,493 --> 00:59:21,273 I'm still having fun at Vast. 936 00:59:21,543 --> 00:59:26,703 And there's lots of interesting stuff still to come. 937 00:59:27,483 --> 00:59:31,593 Having taken a fresh eye to the market. 938 00:59:31,953 --> 00:59:33,753 We got all sorts of good stuff coming. 939 00:59:35,456 --> 00:59:35,906 curtis: Cool. 940 00:59:36,626 --> 00:59:37,046 All right. 941 00:59:37,286 --> 00:59:38,186 I wish you the best. 942 00:59:38,276 --> 00:59:39,786 And thanks Prasanna. 943 00:59:40,786 --> 00:59:43,646 This is one of those cases where your background was very helpful. 944 00:59:43,646 --> 00:59:43,946 I think, 945 00:59:45,178 --> 00:59:47,488 Prasanna: Oh, I try. 946 00:59:47,548 --> 00:59:48,088 I try, 947 00:59:49,108 --> 00:59:56,248 Yeah Yeah. 948 00:59:56,518 --> 00:59:58,678 Having spent a bunch of time building storage arrays. 949 00:59:59,908 --> 01:00:00,688 It helps, but 950 01:00:03,458 --> 01:00:06,488 no, it's still interesting problems though, and, yeah. 951 01:00:07,118 --> 01:00:10,528 Thank you, Howard, for sharing some of the details and indulging in my questions. 952 01:00:10,528 --> 01:00:10,798 So. 953 01:00:11,808 --> 01:00:12,228 Howard: problem