- Events
Monthly Product Demo: Discover the Power of Zilliz Cloud
Webinar
Monthly Product Demo: Discover the Power of Zilliz Cloud
Join the Webinar
Loading...

About this webinar
Join our monthly demo for a technical overview of Zilliz Cloud, a highly scalable and performant vector database service for AI applications
Topics covered
- Zilliz Cloud's scalable architecture
- Key features of the developer-friendly UI
- Security best practices and data privacy
- Highlights from recent product releases
This webinar is an excellent opportunity for developers to learn about Zilliz Cloud's capabilities and how it can support their AI projects. Register now to join our community and stay up-to-date with the latest vector database technology.
WEBVTT
1 00:00:03.825 --> 00:00:04.355 Good morning.
2 00:00:04.385 --> 00:00:06.395 Good. A uh, good afternoon, good evening.
3 00:00:06.425 --> 00:00:07.675 Depends on where you are.
4 00:00:07.845 --> 00:00:10.555 Thank you so much for joining us for today's session.
5 00:00:11.125 --> 00:00:13.995 Today is actually our first monthly, uh,
6 00:00:14.315 --> 00:00:15.435 ZILLS Cloud product demo.
7 00:00:16.015 --> 00:00:18.755 Uh, I'm sfi, I'm a member of he, uh,
8 00:00:18.875 --> 00:00:20.315 a team member here at Zillows.
9 00:00:20.495 --> 00:00:22.875 I'm with my colleague, uh, Amee here.
10 00:00:23.185 --> 00:00:25.675 He's one of the solution architect, right?
11 00:00:25.695 --> 00:00:28.075 That's, that's correct. Your, your title. Amik.
12 00:00:28.075 --> 00:00:29.235 Do you wanna say hi to everyone?
13 00:00:29.965 --> 00:00:33.225 Hi, everybody. Um, great to, uh, great to be here
14 00:00:33.725 --> 00:00:36.625 and, uh, yeah, I'm just a solutions architect
15 00:00:36.625 --> 00:00:39.625 and I'm, um, yeah, looking forward to, uh,
16 00:00:40.045 --> 00:00:42.345 to guiding everyone on, on the, uh,
17 00:00:42.345 --> 00:00:43.665 monthly demo this, this month.
18 00:00:44.495 --> 00:00:46.125 Great. Let's just get started.
19 00:00:46.905 --> 00:00:50.485 Um, so, uh, today I'm very excited to introduce you
20 00:00:50.485 --> 00:00:53.125 to this Zillow Cloud, our scalable
21 00:00:53.385 --> 00:00:55.565 and high performance vector database solution.
22 00:00:58.315 --> 00:01:00.895 Um, try to, okay.
23 00:01:01.115 --> 00:01:05.095 So, uh, there're gonna be, um, two item on the agenda, uh,
24 00:01:05.095 --> 00:01:06.615 since we're gonna have a short webinar.
25 00:01:07.035 --> 00:01:09.855 So the first one, I'm gonna give you guys a high level
26 00:01:10.175 --> 00:01:13.175 overview about, uh, what zills Cloud is.
27 00:01:13.435 --> 00:01:15.495 Um, if you're not super familiar with us,
28 00:01:15.845 --> 00:01:18.295 then amek will give you quick product demo.
29 00:01:22.105 --> 00:01:26.525 Okay. Um, okay.
30 00:01:26.745 --> 00:01:30.885 Uh, as some of you probably know, ULU Cloud is, uh,
31 00:01:30.885 --> 00:01:35.685 built on bu um, so bu is this most famous,
32 00:01:36.025 --> 00:01:37.645 uh, sorry, I wouldn't say this.
33 00:01:37.875 --> 00:01:41.245 Most popular open source, uh, vector database, um,
34 00:01:41.465 --> 00:01:44.325 we donated to Linux Foundation
35 00:01:44.985 --> 00:01:46.925 and back in 2020, um,
36 00:01:46.925 --> 00:01:49.525 because we truly believe building tools
37 00:01:49.595 --> 00:01:51.765 that developers can trust and spec
38 00:01:51.865 --> 00:01:54.525 and contribute to, uh, the numbers here,
39 00:01:54.705 --> 00:01:56.205 it tells quite a bit about the story.
40 00:01:56.745 --> 00:02:00.085 We hit 30 2K plus stars, uh,
41 00:02:00.325 --> 00:02:01.845 I think earlier this year,
42 00:02:02.505 --> 00:02:06.485 and we are seeing more than 68 million docker pools.
43 00:02:06.905 --> 00:02:10.565 So at Core, VUS is a purpose build, um, for
44 00:02:10.675 --> 00:02:13.205 what you're dealing with, purpose build vector to database,
45 00:02:13.685 --> 00:02:14.725 storing, indexing,
46 00:02:14.725 --> 00:02:17.405 and then managing the embedding, um, coming out of
47 00:02:17.405 --> 00:02:20.605 for you the, uh, machine learning model.
48 00:02:21.025 --> 00:02:24.365 Um, whether you are building semantic search recommendation
49 00:02:24.365 --> 00:02:28.205 system or any other, uh, vector search applications,
50 00:02:28.795 --> 00:02:31.725 this is the infrastructure that scales with your needs.
51 00:02:35.945 --> 00:02:39.035 Um, mill was, mill is great. We all love it.
52 00:02:39.135 --> 00:02:43.315 Um, but, uh, uh, we keep hearing from developers and,
53 00:02:43.455 --> 00:02:44.635 and also some, uh,
54 00:02:44.775 --> 00:02:48.355 DevOps people about some pain points in moving application
55 00:02:48.425 --> 00:02:50.475 from prototype into production.
56 00:02:51.015 --> 00:02:53.155 Uh, that's where those cloud come
57 00:02:53.155 --> 00:02:56.635 and would build, uh, three key, like three pillars
58 00:02:56.655 --> 00:02:59.885 to support, um, this, uh, challenges.
59 00:03:00.305 --> 00:03:03.485 So the first one is our, our Cardinal search engine,
60 00:03:03.625 --> 00:03:04.645 if you've ever heard of it.
61 00:03:05.185 --> 00:03:09.405 Um, so, uh, even your VU user, you would love it
62 00:03:09.405 --> 00:03:13.005 because it, uh, provides 10 x faster, um,
63 00:03:13.035 --> 00:03:14.565 performance out of this box.
64 00:03:15.025 --> 00:03:17.285 And we've eliminated the headache
65 00:03:17.305 --> 00:03:19.005 of a manual index tool tuning
66 00:03:19.105 --> 00:03:20.605 and like loss of tuning with her
67 00:03:20.605 --> 00:03:22.125 that is really time consuming
68 00:03:22.145 --> 00:03:24.325 and require lots of, uh, expertise.
69 00:03:25.185 --> 00:03:28.125 The auto index feature basically handles all the
70 00:03:28.165 --> 00:03:31.165 optimization you normally have to figure out yourself.
71 00:03:32.265 --> 00:03:37.245 Um, for the infrastructure, DevOps, uh, folks, we build this
72 00:03:37.305 --> 00:03:39.485 as a true cloud native distributed system.
73 00:03:40.225 --> 00:03:43.365 Uh, this isn't just bu with a cloud wrapper.
74 00:03:43.465 --> 00:03:46.725 Uh, we know like lots of, uh, open source projects,
75 00:03:46.725 --> 00:03:50.365 they just, uh, put their open source, uh, product in,
76 00:03:50.905 --> 00:03:52.845 in a Kubernetes cluster in the cloud
77 00:03:52.905 --> 00:03:54.405 and the code fully managed.
78 00:03:54.905 --> 00:03:55.965 Uh, we don't do this.
79 00:03:56.185 --> 00:03:59.565 Um, villa is actually architected from the ground up, uh,
80 00:03:59.585 --> 00:04:01.845 for horizontal scaling and high availability
81 00:04:02.155 --> 00:04:05.085 because the scale up or down without breaking a suite
82 00:04:05.625 --> 00:04:08.645 and the distributed architecture means better suitability
83 00:04:08.785 --> 00:04:09.925 and the cost efficiency.
84 00:04:10.665 --> 00:04:13.605 So lastly, since we're dealing with production data here,
85 00:04:14.105 --> 00:04:17.565 uh, we take, um, security very, very seriously.
86 00:04:18.145 --> 00:04:20.565 Um, secure both security and reliability.
87 00:04:21.095 --> 00:04:24.205 We're talking about, um, the better tested performance
88 00:04:24.205 --> 00:04:26.165 that meet enterprise security requirements
89 00:04:26.275 --> 00:04:27.885 because nobody wants to explain
90 00:04:27.885 --> 00:04:31.525 to their CISO why they choose a solution without
91 00:04:31.565 --> 00:04:32.965 proper security controls.
92 00:04:35.895 --> 00:04:37.155 Um, so let's, uh,
93 00:04:37.155 --> 00:04:39.235 dig a little bit deeper into the technical details
94 00:04:39.295 --> 00:04:40.755 of the cardio search engine.
95 00:04:41.375 --> 00:04:44.235 If you've worked with Novis before, you are recognized.
96 00:04:44.235 --> 00:04:46.755 Some common pain, uh, we are addressing here.
97 00:04:47.005 --> 00:04:48.635 First is indexing.
98 00:04:49.415 --> 00:04:52.915 We are, we all know the drill spending hours,
99 00:04:53.455 --> 00:04:56.115 tuning parameters, running benchmarks,
100 00:04:56.415 --> 00:05:00.035 and still wondering if we've hit the optimal performance.
101 00:05:00.495 --> 00:05:03.155 The auto index system basically handles this for you.
102 00:05:03.615 --> 00:05:05.355 It analyze your dataset
103 00:05:05.575 --> 00:05:06.915 and whatever, uh,
104 00:05:07.235 --> 00:05:10.635 hardware underlying online hardware then picks the most
105 00:05:10.915 --> 00:05:11.915 efficient search strategy.
106 00:05:12.295 --> 00:05:14.435 So there's no more guesswork over there.
107 00:05:15.135 --> 00:05:17.155 The core architecture is pretty straightforward
108 00:05:17.665 --> 00:05:19.875 algorithm layer with index building
109 00:05:19.935 --> 00:05:22.955 and a searching percolation layer for quantization
110 00:05:23.015 --> 00:05:27.035 and a refinement and a storage using a graph IVF hybrid
111 00:05:27.395 --> 00:05:29.755 approach that is actually really unique in the industry.
112 00:05:30.185 --> 00:05:32.675 Most of our competitors probably either, no,
113 00:05:32.675 --> 00:05:34.955 they're pretty much just used like IVF, um,
114 00:05:35.015 --> 00:05:38.675 but this kind of IVF, um, sorry, the mostly used graph.
115 00:05:39.055 --> 00:05:42.355 Um, but this type of graph IVF hybrid is, uh, interesting
116 00:05:42.455 --> 00:05:45.075 unless you tune the trade off between the capacity
117 00:05:45.135 --> 00:05:47.075 and performance based on your workloads.
118 00:05:47.495 --> 00:05:50.675 In our tests, we've getting about 50% more capacity
119 00:05:50.735 --> 00:05:55.435 and seeing search speeds up to 10 x faster than, uh, vis.
120 00:05:56.285 --> 00:06:00.135 Um, we've also done a bunch of low level optimization, uh,
121 00:06:00.375 --> 00:06:01.735 kernel level stuff and CPU
122 00:06:01.735 --> 00:06:03.655 and metrics analysis we could squeeze,
123 00:06:04.195 --> 00:06:05.415 uh, per bad performance.
124 00:06:07.105 --> 00:06:10.125 So let's talk a little bit more about cloud Cloud native
125 00:06:10.505 --> 00:06:11.525 and fully managed.
126 00:06:11.755 --> 00:06:14.045 What, what that actually means in practice.
127 00:06:14.785 --> 00:06:18.045 Uh, we put like comparison here, just like for you
128 00:06:18.045 --> 00:06:19.085 to easy understand.
129 00:06:19.625 --> 00:06:23.885 Um, you see, um, we, what we build a,
130 00:06:23.885 --> 00:06:26.485 those cloud code like fully managed rather than just
131 00:06:26.485 --> 00:06:27.605 like a hosted solution.
132 00:06:28.265 --> 00:06:32.125 Um, so on the most left call column, um, you can see, um,
133 00:06:32.185 --> 00:06:34.045 that's pretty much like open source, right?
134 00:06:34.465 --> 00:06:35.965 Anyone who runs VIS
135 00:06:36.025 --> 00:06:38.325 or any other open source vector database, um,
136 00:06:38.775 --> 00:06:40.965 knows there are like lots of operational overhead.
137 00:06:41.585 --> 00:06:44.605 Um, the open source version is great for development,
138 00:06:44.705 --> 00:06:46.885 we would say, but when you hit a production,
139 00:06:47.145 --> 00:06:49.685 you're handling everything yourself from data migration
140 00:06:49.825 --> 00:06:52.605 to scaling to upgrades, uh,
141 00:06:52.605 --> 00:06:55.045 with some hosted solution available on the market.
142 00:06:55.345 --> 00:06:58.845 Um, you get an infrastructure and Kubernetes management, um,
143 00:06:58.865 --> 00:07:00.365 but you're still on the hook for most
144 00:07:00.365 --> 00:07:01.605 of the operation stuff still.
145 00:07:02.185 --> 00:07:04.805 Uh, what's different from our management approach is
146 00:07:04.835 --> 00:07:07.325 that we handle all those production headaches.
147 00:07:07.665 --> 00:07:10.885 So there's a zero downtime scaling, you can scale up
148 00:07:11.265 --> 00:07:13.765 or down with without service interruption.
149 00:07:13.865 --> 00:07:15.045 We think that's super important.
150 00:07:15.665 --> 00:07:18.245 Um, automatic backup and data migration.
151 00:07:18.465 --> 00:07:20.605 So you don't need to write customer scripts
152 00:07:20.865 --> 00:07:21.925 or manage this process.
153 00:07:22.825 --> 00:07:25.325 Um, there's a performance tuning based on
154 00:07:25.325 --> 00:07:26.445 your specific use case.
155 00:07:26.745 --> 00:07:29.565 So we optimize based on your actual workload pattern.
156 00:07:29.745 --> 00:07:32.935 Uh, um, we will share that a bit more later. Course.
157 00:07:32.935 --> 00:07:34.815 So there's auto scaling, um,
158 00:07:35.235 --> 00:07:37.175 and a hands off update and patching.
159 00:07:37.315 --> 00:07:38.615 So the security patches
160 00:07:38.615 --> 00:07:41.415 and the version upgrades that happen, uh, just like
161 00:07:41.415 --> 00:07:43.255 behind the scenes, you don't have to worry about it.
162 00:07:44.045 --> 00:07:46.375 This essentially means your team can focus on building
163 00:07:46.775 --> 00:07:48.575 features instead of managing the infrastructure.
164 00:07:50.325 --> 00:07:53.465 So, um, a big part of the cloud native is also about how
165 00:07:53.465 --> 00:07:54.545 to manage multi-tenancy.
166 00:07:54.605 --> 00:07:57.745 Um, amik, uh, do you want to explain
167 00:07:57.885 --> 00:07:59.545 how our multi-tenancy work here?
168 00:08:01.675 --> 00:08:02.965 Yeah, absolutely.
169 00:08:03.025 --> 00:08:07.205 So we have three, uh, layers, uh, a three layered approach,
170 00:08:07.385 --> 00:08:09.045 uh, when it comes to multi-tenancy.
171 00:08:09.585 --> 00:08:12.925 Um, the first is database oriented multi-tenancy.
172 00:08:13.505 --> 00:08:15.245 The second is collection oriented,
173 00:08:15.905 --> 00:08:18.445 and then the third is partition oriented.
174 00:08:19.225 --> 00:08:23.085 Um, what you see right now, uh, describes our parti.
175 00:08:23.085 --> 00:08:25.045 Uh, the diagram that you see right now describes our
176 00:08:25.045 --> 00:08:28.125 partition oriented, uh, multi-tenancy,
177 00:08:28.685 --> 00:08:31.325 specifically our partition key based multi-tenancy.
178 00:08:32.225 --> 00:08:35.965 And upon, uh, the creation of a collection, um,
179 00:08:35.985 --> 00:08:39.005 you can actually nominate a tenant field
180 00:08:39.745 --> 00:08:42.285 and make that your partition key field.
181 00:08:43.065 --> 00:08:44.325 And so VIS
182 00:08:44.825 --> 00:08:49.205 or Zillow will basically store entities in a partition
183 00:08:49.205 --> 00:08:52.845 according to the hash value of the partition key field
184 00:08:53.185 --> 00:08:54.885 and only search the partition that
185 00:08:54.950 --> 00:08:56.805 that contains the specific partition key.
186 00:08:57.665 --> 00:09:00.925 And the, the reason why this strategy is so effective is
187 00:09:00.925 --> 00:09:04.685 because it lifts the limit on the maximum number of tenants
188 00:09:04.685 --> 00:09:06.445 that a Novus collection can support.
189 00:09:07.105 --> 00:09:09.765 Uh, for example, the, the, the maximum number of tenants,
190 00:09:10.505 --> 00:09:14.565 um, for a collection is, uh, less than 10,000.
191 00:09:15.025 --> 00:09:18.485 But with the partition key based approach, the, the number
192 00:09:18.505 --> 00:09:22.085 of tenants that can be supported, um, exceeds 10 million.
193 00:09:23.005 --> 00:09:25.445 Additionally, with a partition key based multi-tenancy
194 00:09:25.645 --> 00:09:29.045 approach, it is a lot faster than using, um, for example,
195 00:09:29.185 --> 00:09:30.445 one collection per tenant
196 00:09:31.065 --> 00:09:33.725 or having one collection for all tenants
197 00:09:33.725 --> 00:09:36.605 and then using a filtering system, um, for every,
198 00:09:37.105 --> 00:09:38.245 um, every partition.
199 00:09:38.865 --> 00:09:41.005 And so it's the most, uh,
200 00:09:41.005 --> 00:09:44.285 dynamic offering we have when it comes to scalability
201 00:09:44.545 --> 00:09:46.085 as well as, uh, speed.
202 00:09:48.055 --> 00:09:51.765 Cool. Um, then let's continue
203 00:09:51.765 --> 00:09:54.485 to talk a little bit about, uh, the enterprise feature.
204 00:09:55.185 --> 00:09:58.165 Uh, security is obviously, uh, critical when you are dealing
205 00:09:58.165 --> 00:09:59.925 with product production data.
206 00:10:00.545 --> 00:10:02.525 We build this with a multilayered approach.
207 00:10:02.785 --> 00:10:04.165 Uh, there's authentications
208 00:10:04.165 --> 00:10:08.045 through SSO access control at a multiple levels, um,
209 00:10:08.345 --> 00:10:09.685 or back pri private
210 00:10:09.865 --> 00:10:12.005 and points to make sure you are never exposed
211 00:10:12.005 --> 00:10:13.165 to your data to the internet.
212 00:10:13.585 --> 00:10:15.445 And everything is encrypted end to end,
213 00:10:15.445 --> 00:10:17.165 both in transit and at rest.
214 00:10:17.665 --> 00:10:19.085 Uh, on the deployment side,
215 00:10:19.625 --> 00:10:22.485 you can see we've structured this as a way to,
216 00:10:22.485 --> 00:10:25.205 that give you flexibility whether you wanna run on
217 00:10:25.245 --> 00:10:28.205 A-W-S-G-C-P Azure or bring your cloud.
218 00:10:28.415 --> 00:10:30.645 We're gonna talk about that, that a little bit more later.
219 00:10:31.025 --> 00:10:32.245 Uh, it's all supported.
220 00:10:32.625 --> 00:10:34.525 Uh, the key thing here is that if you need
221 00:10:34.525 --> 00:10:37.205 to keep data within your VPC, you can
222 00:10:37.815 --> 00:10:40.445 check out this architecture diagram on the right.
223 00:10:40.625 --> 00:10:43.005 Uh, each customer get physical isolated cluster,
224 00:10:43.545 --> 00:10:44.645 no shared resources.
225 00:10:44.955 --> 00:10:47.605 Your service are completely separated from other customers.
226 00:10:48.305 --> 00:10:50.805 Um, the purple layer at the bottom, that's, uh,
227 00:10:50.805 --> 00:10:52.805 where we handle all your data storage
228 00:10:53.075 --> 00:10:54.365 with building encryption.
229 00:10:54.985 --> 00:10:57.245 For those of you worried about uptime, uh,
230 00:10:57.255 --> 00:10:59.725 we're running multi AZ deployment with, uh,
231 00:10:59.725 --> 00:11:02.445 99.95% SOA.
232 00:11:02.585 --> 00:11:05.085 That's uptime, SOA building backup
233 00:11:05.305 --> 00:11:08.365 and disaster recovery, uh, come, come standards.
234 00:11:10.085 --> 00:11:13.545 Uh, so let's talk a bit more about, uh, BIOC.
235 00:11:13.635 --> 00:11:16.025 We've heard like lots of customer feedback about keep their
236 00:11:16.025 --> 00:11:19.785 data sovereignty, also minimize operational burden.
237 00:11:20.285 --> 00:11:22.145 So we released BIOC last year.
238 00:11:22.425 --> 00:11:25.945 Actually today we announced another really big upgrade,
239 00:11:26.205 --> 00:11:27.545 uh, to RBLC.
240 00:11:27.575 --> 00:11:28.785 I'll put the link a little bit later
241 00:11:28.845 --> 00:11:30.665 so you can check if you're interested in this,
242 00:11:30.725 --> 00:11:32.145 uh, deployment model.
243 00:11:32.925 --> 00:11:35.265 So, uh, to meet our customer's needs,
244 00:11:35.405 --> 00:11:37.665 we offer our bring all on cloud option.
245 00:11:38.215 --> 00:11:39.865 This is for team who need
246 00:11:39.865 --> 00:11:42.825 to keep everything within their own VPC usually due
247 00:11:42.845 --> 00:11:44.465 to compliance requirements
248 00:11:45.005 --> 00:11:49.305 or internal security policy, it's available, um, AWS
249 00:11:49.305 --> 00:11:52.585 and A GCP right now with Azure Support coming soon.
250 00:11:53.185 --> 00:11:55.405 Uh, of course on the left hand we still,
251 00:11:55.665 --> 00:11:58.845 you can see we still, um, got our fully managed service,
252 00:11:58.955 --> 00:12:01.565 basically vis Ray engineer specifically
253 00:12:01.665 --> 00:12:03.285 for cloud native deployment.
254 00:12:03.745 --> 00:12:06.285 You can run this on A-W-S-G-C-P or Azure.
255 00:12:06.465 --> 00:12:07.725 We handle everything for you.
256 00:12:07.795 --> 00:12:12.605 Scaling, upgrade, updates, maintenance, uh, so, um,
257 00:12:12.785 --> 00:12:15.765 that's, uh, that's, that's the Zillow Cloud cell service.
258 00:12:17.555 --> 00:12:20.255 So let me talk a little bit more about how
259 00:12:20.965 --> 00:12:25.095 exactly our BIOC uh, make the workload secure.
260 00:12:25.575 --> 00:12:28.985 'cause, uh, we heard, uh, BIOC is, uh, like, uh,
261 00:12:29.165 --> 00:12:31.345 not something new, but like many,
262 00:12:31.345 --> 00:12:33.145 many vendor actually didn't do the correct.
263 00:12:33.645 --> 00:12:35.825 Uh, so we want to share with you our approach.
264 00:12:36.125 --> 00:12:38.945 So this diagram just shows like, uh, our bureau
265 00:12:39.505 --> 00:12:42.065 security architecture with two distinct environment,
266 00:12:42.555 --> 00:12:45.425 those cloud VPC and the customer manage VPC.
267 00:12:45.925 --> 00:12:49.345 So on the left, uh, it says those Cloud VPC,
268 00:12:49.725 --> 00:12:52.905 it contains our control panel, uh, the BLC controller
269 00:12:52.905 --> 00:12:53.985 for managing resources
270 00:12:54.365 --> 00:12:56.585 and a system monitor for tracking performance.
271 00:12:57.205 --> 00:13:00.305 Um, on the right, this is the customer manage VPC.
272 00:13:00.565 --> 00:13:04.225 It contains the deployment running Kubernetes clusters along
273 00:13:04.225 --> 00:13:05.305 with the monetary tools.
274 00:13:06.045 --> 00:13:09.625 So you see the arrow labeled, uh, outbound 4, 4 3.
275 00:13:09.925 --> 00:13:12.825 So, um, we did this for you.
276 00:13:12.925 --> 00:13:15.625 Uh, so everything is HT DP encrypted.
277 00:13:16.125 --> 00:13:17.585 Uh, so most importantly,
278 00:13:17.655 --> 00:13:20.745 only the customer side can initiate this connection.
279 00:13:20.855 --> 00:13:22.905 Nothing can reach from the outside.
280 00:13:23.765 --> 00:13:28.105 So, um, that's making it like a, um, really secure.
281 00:13:28.975 --> 00:13:31.785 Then when you need to manage resources like scale
282 00:13:31.785 --> 00:13:35.225 and vis, it happens through the strict limited permission,
283 00:13:35.725 --> 00:13:38.425 um, that's actually newly, newly released system
284 00:13:38.695 --> 00:13:40.065 that the customer control.
285 00:13:40.605 --> 00:13:44.545 Um, so, um, think of like a giving a contract specific keys
286 00:13:44.625 --> 00:13:45.705 to the specific rooms.
287 00:13:46.005 --> 00:13:48.265 So not amo key to the buildings.
288 00:13:48.325 --> 00:13:49.785 So that's how we make it safe.
289 00:13:50.205 --> 00:13:53.145 So this kind of a combination of outbound only communication
290 00:13:53.245 --> 00:13:54.385 and a controlled permission
291 00:13:54.615 --> 00:13:58.545 that makes B-R-B-B-I-O-C truly enterprise grade secure.
292 00:14:01.135 --> 00:14:04.315 Um, so, um, very quickly actually, we, uh,
293 00:14:04.715 --> 00:14:07.875 released some new version of mil, uh, ulus.
294 00:14:08.135 --> 00:14:09.715 Uh, I think last week.
295 00:14:10.415 --> 00:14:13.435 Um, the most important feature this time we released is
296 00:14:13.675 --> 00:14:16.595 actually VU 2.5 is available on ulus
297 00:14:16.595 --> 00:14:17.755 Cloud as a public preview.
298 00:14:18.145 --> 00:14:21.875 There's like lots of great features, uh, for text search,
299 00:14:22.145 --> 00:14:25.755 text matching, bit map index, and there's more, uh, hours.
300 00:14:25.985 --> 00:14:28.955 I'll put the, uh, release notes, uh, for,
301 00:14:28.975 --> 00:14:30.035 for you to review later.
302 00:14:30.575 --> 00:14:34.235 Uh, but the most most exciting one is the full text search.
303 00:14:34.815 --> 00:14:36.955 So, Amik, do you want to share a bit more
304 00:14:37.055 --> 00:14:38.835 how exactly our text search work?
305 00:14:41.665 --> 00:14:42.755 Yeah, absolutely.
306 00:14:42.895 --> 00:14:46.075 Um, so the principle behind, uh, full text search is
307 00:14:46.075 --> 00:14:49.835 to first input, um, your raw text data into the,
308 00:14:50.225 --> 00:14:51.295 into the platform.
309 00:14:51.755 --> 00:14:55.535 And then, um, you run it through an analyzer that, uh,
310 00:14:55.535 --> 00:14:59.295 basically generate that, uh, that breaks down the tokens
311 00:14:59.715 --> 00:15:02.215 of those, uh, of the raw text.
312 00:15:02.995 --> 00:15:06.535 And then afterwards, uh, we feed those tokens, um,
313 00:15:06.925 --> 00:15:11.655 into a function that then converts those, um,
314 00:15:12.265 --> 00:15:16.755 words into a sparse embedding.
315 00:15:17.555 --> 00:15:20.135 And once that occurs, um,
316 00:15:20.315 --> 00:15:22.855 we then group those sparse embeddings, um,
317 00:15:22.855 --> 00:15:25.615 within a collection based off of any filtering
318 00:15:25.615 --> 00:15:27.375 or search that occurs, um,
319 00:15:27.435 --> 00:15:32.095 and then use that, uh, uh, to put that into our BM 25, uh,
320 00:15:32.095 --> 00:15:36.775 scoring algorithm to, um, generate a certain top K results.
321 00:15:37.555 --> 00:15:39.415 Um, and then based off of the results
322 00:15:39.415 --> 00:15:43.975 that the user receives, they can then, um, sort of mod
323 00:15:44.275 --> 00:15:46.215 and iterate over the, the text queries
324 00:15:46.215 --> 00:15:48.175 that they insert into the analyzer.
325 00:15:48.755 --> 00:15:51.575 And, um, with that feedback loop generate, uh,
326 00:15:51.575 --> 00:15:52.575 better results over time.
327 00:15:55.555 --> 00:16:00.525 Cool. Um, then let's get into the more exciting demo part.
328 00:16:00.625 --> 00:16:02.285 Uh, Amik, do you wanna share your screen?
329 00:16:03.075 --> 00:16:04.005 Yeah, absolutely.
330 00:16:06.685 --> 00:16:08.385 Um, great.
331 00:16:08.805 --> 00:16:13.675 So just going into the, so this is the Zillow cloud, uh,
332 00:16:14.055 --> 00:16:15.235 UI and platform itself.
333 00:16:15.965 --> 00:16:19.145 And so on the left hand side, you can, you have a section
334 00:16:19.145 --> 00:16:21.145 where you can, you know, create your clusters.
335 00:16:21.885 --> 00:16:25.945 Um, you can also manage backups, uh,
336 00:16:26.115 --> 00:16:29.185 migrations from different, uh, platforms.
337 00:16:29.285 --> 00:16:33.625 So we support, uh, migrations for you from VIS to Zillow,
338 00:16:34.125 --> 00:16:36.065 um, within different organizations.
339 00:16:36.845 --> 00:16:40.745 Um, and so the, the overarching motive of the,
340 00:16:40.765 --> 00:16:44.225 of the features we provide is to make the Zillow cloud
341 00:16:44.745 --> 00:16:46.545 platform as simple to use as possible
342 00:16:46.565 --> 00:16:48.225 and to reduce the amount of work
343 00:16:49.025 --> 00:16:51.705 required on your team's end in order
344 00:16:51.885 --> 00:16:53.985 to start using our platform
345 00:16:54.405 --> 00:16:57.985 and as well as maintain the, the, the compute units
346 00:16:57.985 --> 00:17:00.505 that you guys spin up on our platform as well.
347 00:17:00.645 --> 00:17:03.505 So hence the reason why we support, uh,
348 00:17:03.505 --> 00:17:05.585 migrations from many external data sources.
349 00:17:06.405 --> 00:17:10.425 Um, also from our existing organizations, we give you a UI
350 00:17:10.445 --> 00:17:14.185 to manage your jobs, manage collaborators, um,
351 00:17:14.285 --> 00:17:16.225 and even set alerts, um,
352 00:17:16.525 --> 00:17:19.585 to manage when data is imported or exported.
353 00:17:19.885 --> 00:17:23.865 And then finally, um, as Steff was mentioning, uh, we,
354 00:17:24.005 --> 00:17:26.105 we allow you to create a private endpoint as well
355 00:17:26.105 --> 00:17:27.865 as manage certain networks.
356 00:17:27.865 --> 00:17:30.585 And then finally, we offer integrations with Datadog,
357 00:17:30.585 --> 00:17:35.185 Prometheus and, um, S3 for, um, allowing your team
358 00:17:35.205 --> 00:17:37.465 to manage your logs in one centralized location
359 00:17:38.085 --> 00:17:41.665 or to push backup files to to Amazon S3.
360 00:17:42.615 --> 00:17:44.875 Um, I'd like to now dive deeper into
361 00:17:45.175 --> 00:17:46.435 how, how to create a cluster.
362 00:17:47.175 --> 00:17:50.645 Um, and so we do offer three plans, uh, the free plan,
363 00:17:50.645 --> 00:17:52.005 serverless and, and dedicated.
364 00:17:52.315 --> 00:17:55.205 Most of our enterprise customers go with our dedicated plan.
365 00:17:55.545 --> 00:17:59.005 And just today we actually released the, uh, the ability to,
366 00:17:59.145 --> 00:18:01.445 for, um, users on your end
367 00:18:01.445 --> 00:18:05.005 to spin up A BYO cluster BYOC cluster on their own.
368 00:18:05.725 --> 00:18:09.285 Previously it required an it, um, uh, an it, uh,
369 00:18:09.285 --> 00:18:12.165 individual contributor to manage that, manage that system.
370 00:18:12.625 --> 00:18:16.205 Um, you can name your cluster, um, for d for, uh, free
371 00:18:16.205 --> 00:18:19.525 and serverless CL clusters, um, only GCPU supported,
372 00:18:19.525 --> 00:18:22.285 but for dedicated, we have all three CLO major cloud
373 00:18:22.285 --> 00:18:24.525 supported as well as these existing regions.
374 00:18:25.495 --> 00:18:28.075 Um, we also have three CU types as well.
375 00:18:28.215 --> 00:18:31.755 So performance optimized is for your highest, uh,
376 00:18:31.775 --> 00:18:33.875 for your lowest latency and highest throughput cases.
377 00:18:34.455 --> 00:18:36.915 Um, capacity optimized is for still,
378 00:18:36.915 --> 00:18:38.195 you get a very good LA latency.
379 00:18:38.195 --> 00:18:39.875 You get approximately a hundred millisecond latency,
380 00:18:39.935 --> 00:18:43.275 but um, it's has more enhanced storage capabilities.
381 00:18:43.295 --> 00:18:45.435 And then ex finally, extended capacity
382 00:18:46.135 --> 00:18:49.355 is your most affordable solution with, um,
383 00:18:49.535 --> 00:18:52.715 sub 500 milli millisecond to sub one second latency,
384 00:18:52.715 --> 00:18:56.675 but the CU can store, um, store many more vectors.
385 00:18:57.395 --> 00:19:01.175 Um, so going back now to this is an existing cluster
386 00:19:01.175 --> 00:19:03.215 that I've already created.
387 00:19:04.155 --> 00:19:07.695 Um, so here we can, uh, manage our cluster details
388 00:19:08.165 --> 00:19:09.855 with the CU size and plan.
389 00:19:10.195 --> 00:19:12.375 So you can upgrade your plan here if you'd like.
390 00:19:12.995 --> 00:19:16.135 Um, you can scale your CU as well manually.
391 00:19:16.755 --> 00:19:20.705 Um, and then here you can manage your collections.
392 00:19:21.285 --> 00:19:24.465 Um, and so collections, um, can be used either
393 00:19:24.465 --> 00:19:26.625 for multi-tenancy as I, me mentioned before,
394 00:19:27.205 --> 00:19:29.785 or you can create a specific collection on your own with,
395 00:19:29.805 --> 00:19:30.905 you know, specific fields.
396 00:19:31.005 --> 00:19:32.985 So here you can define the schema.
397 00:19:33.445 --> 00:19:35.545 Um, and then within your advanced things as well,
398 00:19:35.545 --> 00:19:36.665 you can create dynamic fields.
399 00:19:36.665 --> 00:19:39.305 And so let's say you create a schema, um,
400 00:19:39.925 --> 00:19:41.505 you load your data into the collection,
401 00:19:41.765 --> 00:19:44.905 but then you want to add, uh, another, um,
402 00:19:45.405 --> 00:19:46.585 column to that schema.
403 00:19:46.585 --> 00:19:48.625 You can add a dynamic di dynamic field as well.
404 00:19:49.005 --> 00:19:51.745 And then if you're, or if you're looking to, um,
405 00:19:52.255 --> 00:19:53.345 perform partition key
406 00:19:53.345 --> 00:19:54.745 multi-tenancy, you can enable that here.
407 00:19:55.565 --> 00:19:56.705 Um, as well,
408 00:20:00.405 --> 00:20:01.745 the metrics tab, uh,
409 00:20:02.395 --> 00:20:04.305 tells you about the health of your, uh, cu.
410 00:20:04.645 --> 00:20:08.585 So either your computation, um, com computational capacity
411 00:20:09.045 --> 00:20:11.665 or the, your storage capacity for how many, um,
412 00:20:11.815 --> 00:20:14.625 vectors have been stored, um, et cetera.
413 00:20:15.125 --> 00:20:17.105 And then with your users as well, this is
414 00:20:17.105 --> 00:20:18.385 where you can perform a little bit
415 00:20:18.385 --> 00:20:22.145 of role-based access control to see, um, who has control.
416 00:20:23.065 --> 00:20:26.125 Um, so you can, for example, the multiple, uh,
417 00:20:26.185 --> 00:20:28.965 you can select, uh, for different user roles here as well.
418 00:20:30.095 --> 00:20:35.065 Um, you can, um, so then, uh, this is at more at the,
419 00:20:35.165 --> 00:20:36.985 um, collection level versus
420 00:20:36.985 --> 00:20:38.145 this is more at the cluster level.
421 00:20:38.205 --> 00:20:43.185 So here you can assign different privileges, um,
422 00:20:43.605 --> 00:20:46.145 for different, uh, le levels of granularity.
423 00:20:46.565 --> 00:20:50.645 Um, for, and I think this slide, um, describes it best.
424 00:20:50.825 --> 00:20:54.845 So, um, at the organization level, um,
425 00:20:54.985 --> 00:20:57.685 you can assign privileges as well as at the project level.
426 00:20:57.825 --> 00:21:00.045 So that's really what's stored in the control plane.
427 00:21:00.265 --> 00:21:02.085 And then within the data plane for customers
428 00:21:02.085 --> 00:21:03.965 who are sensitive about who accesses their data,
429 00:21:04.265 --> 00:21:08.405 we offer cluster level roles, um, which is what, uh,
430 00:21:08.455 --> 00:21:10.045 which is what you were seeing in the ui,
431 00:21:10.045 --> 00:21:11.125 at the database level,
432 00:21:11.185 --> 00:21:13.605 and at the, at the collection level as well.
433 00:21:15.545 --> 00:21:19.005 Um, and so you can see that within a cluster, um,
434 00:21:19.185 --> 00:21:20.845 you can select between database privileges
435 00:21:20.845 --> 00:21:21.925 as well as cluster privileges.
436 00:21:21.925 --> 00:21:25.085 And then even you have, you know, read only read ReadWrite
437 00:21:25.185 --> 00:21:27.965 or admin privileges.
438 00:21:28.585 --> 00:21:33.255 Um, and then finally, um, in this screen is where you can,
439 00:21:33.275 --> 00:21:34.495 uh, generate your backups.
440 00:21:34.495 --> 00:21:36.935 So you can schedule automatic backups as well for your data
441 00:21:37.315 --> 00:21:38.775 or schedule manual backups.
442 00:21:38.995 --> 00:21:43.295 And you can even push these backups into your, um, S3 bucket
443 00:21:43.565 --> 00:21:46.545 with our, um, with our integrations.
444 00:21:47.765 --> 00:21:52.045 Um, and then within a collection itself.
445 00:21:52.185 --> 00:21:54.245 So if we click into a collection,
446 00:21:54.905 --> 00:21:58.795 what we can see is we can also offer a way to manage, um,
447 00:21:59.555 --> 00:22:00.995 a each individual collection.
448 00:22:01.015 --> 00:22:03.835 So here you can see the schema that's been created, um,
449 00:22:03.835 --> 00:22:05.995 with the field types and, and the values.
450 00:22:06.655 --> 00:22:08.675 Um, here you can import your data.
451 00:22:09.215 --> 00:22:11.755 Um, so if you'd like to insert your data, we'll, you know,
452 00:22:11.825 --> 00:22:15.975 provide you with the, uh, API code you need in order to, um,
453 00:22:16.435 --> 00:22:17.775 you know, perform that seamlessly.
454 00:22:18.355 --> 00:22:21.195 Um, you can also see a preview of your data.
455 00:22:21.255 --> 00:22:23.115 So here you can see your vectors listed as well
456 00:22:23.115 --> 00:22:24.315 as all the fields necessary.
457 00:22:25.055 --> 00:22:27.395 And then finally, you can perform your vector search
458 00:22:27.755 --> 00:22:29.315 actually as well within the platform.
459 00:22:29.695 --> 00:22:32.855 So you can, um, put your query vector here, um,
460 00:22:34.395 --> 00:22:36.775 modify your top, give value, or even add a filter.
461 00:22:37.395 --> 00:22:39.655 So a, a filter is also another way
462 00:22:39.655 --> 00:22:42.655 to perform multi-tenancy at the collection level, um,
463 00:22:42.755 --> 00:22:44.815 to specify a filtering collection.
464 00:22:44.815 --> 00:22:47.575 However, we strongly recommend using partition keys
465 00:22:47.915 --> 00:22:49.255 to perform multi-tenancy simply
466 00:22:49.255 --> 00:22:52.775 because the performance is a lot superior, um, on our,
467 00:22:52.775 --> 00:22:55.255 an algorithms while using a partition key based
468 00:22:55.255 --> 00:22:58.055 multi-tenancy approach versus a collection level, um,
469 00:22:58.205 --> 00:23:00.495 filtering based multi-tenancy approach.
470 00:23:01.555 --> 00:23:03.815 Um, so here you can actually perform a vector search
471 00:23:03.955 --> 00:23:07.135 as well, um, and you can see, uh, the results.
472 00:23:07.395 --> 00:23:11.455 Um, uh, yeah, so I think that, um,
473 00:23:11.845 --> 00:23:16.245 that concludes the, that concludes the demo on, um,
474 00:23:16.545 --> 00:23:18.125 on my end, I think.
475 00:23:20.855 --> 00:23:22.945 Cool. Is there any, uh, questions?
476 00:23:23.205 --> 00:23:25.745 Uh, so you can paste it on the q and a.
477 00:23:26.405 --> 00:23:29.065 Uh, actually another team member, John, who's our head of,
478 00:23:29.065 --> 00:23:31.145 uh, dev Rail is here, uh, as well.
479 00:23:31.565 --> 00:23:35.625 So, uh, if you have any questions regarding the demo
480 00:23:36.245 --> 00:23:40.945 or product or anything, um, those related, uh,
481 00:23:41.135 --> 00:23:42.425 feel free to shoot it to us.
482 00:23:43.325 --> 00:23:45.330 So this is our first time to do the demo.
483 00:23:45.985 --> 00:23:48.525 If you have any feedback, um,
484 00:23:49.305 --> 00:23:51.045 we would really appreciate that as well.
485 00:23:51.045 --> 00:23:55.885 I'll just like shoot it our way. Okay.
486 00:23:55.985 --> 00:23:58.845 Uh, it seems we did a great job of explaining everything.
487 00:23:58.845 --> 00:24:00.405 There's no question for now,
488 00:24:00.425 --> 00:24:03.765 but if you can think of any down the road, uh, feel free
489 00:24:03.765 --> 00:24:08.525 to join our, uh, oh, there's, uh, planning regular training.
490 00:24:09.345 --> 00:24:11.765 Uh, what do you mean by regular training?
491 00:24:11.765 --> 00:24:14.285 Similar to Google, we're, we're gonna do this type
492 00:24:14.285 --> 00:24:15.645 of a demo, um, monthly.
493 00:24:16.225 --> 00:24:19.205 We actually also have like a lots of, uh, webinar, uh,
494 00:24:19.345 --> 00:24:21.685 to have a different kind of, uh, topic.
495 00:24:21.875 --> 00:24:24.245 Some of them are more hands-on, some
496 00:24:24.245 --> 00:24:27.285 of them are talking about some like a trend hearing from
497 00:24:27.385 --> 00:24:31.045 how the other, uh Oh, okay.
498 00:24:31.605 --> 00:24:33.045 I don't think we'll have like a formal
499 00:24:33.195 --> 00:24:34.445 kind of program for now.
500 00:24:34.585 --> 00:24:38.005 Um, but, uh, there's still like lots of resources if you go
501 00:24:38.005 --> 00:24:41.365 to our, uh, website Learn,
502 00:24:41.745 --> 00:24:43.725 and those are come slash learn
503 00:24:44.065 --> 00:24:46.245 and there are like lots of good resources over there.
504 00:24:50.955 --> 00:24:54.645 Okay. So, uh, one, oh, I think there's one question here.
505 00:24:55.785 --> 00:24:56.845 Um, so Sergo,
506 00:24:56.905 --> 00:24:59.925 if you are interested in knowing more about ZI
507 00:25:00.145 --> 00:25:03.365 and vus, uh, feel free to shoot us a, a message.
508 00:25:03.505 --> 00:25:06.645 We can, you know, if that's, um, just ad hoc, you want
509 00:25:06.645 --> 00:25:09.525 to learn something about the specific like features
510 00:25:09.865 --> 00:25:13.005 or you have a particular use case, uh, you can, um,
511 00:25:13.095 --> 00:25:14.765 reach out to us through support
512 00:25:15.345 --> 00:25:19.125 or if you have a open source vu use case, we have the Vu,
513 00:25:19.465 --> 00:25:22.285 uh, office hour being scheduled twice every week.
514 00:25:22.745 --> 00:25:27.315 Um, so you can check out the, uh, mill.io website, uh,
515 00:25:28.135 --> 00:25:31.715 for the meals office hour for this cloud, you can reach out
516 00:25:31.715 --> 00:25:32.835 to z cloud.com.
517 00:25:32.835 --> 00:25:34.195 There's a support page.
518 00:25:34.335 --> 00:25:37.555 You can reach out to us through written format. Thank you.
519 00:25:38.505 --> 00:25:41.645 Uh, I think my young, you have a, you want to, yeah,
520 00:25:41.645 --> 00:25:42.925 you can ask a question live.
521 00:25:42.925 --> 00:25:47.425 That's fine. Yeah. Do you wanna talk?
522 00:25:47.625 --> 00:25:49.425 I think you can unmute yourself and talk. Yeah,
523 00:25:50.215 --> 00:25:51.215 Yeah. Uh, I didn't
524 00:25:51.215 --> 00:25:53.385 have any question as of now. Okay.
525 00:25:53.685 --> 00:25:57.785 But, uh, once I will learn, uh, from the session
526 00:25:58.135 --> 00:26:01.785 that you have, and, uh, then I will, uh,
527 00:26:02.535 --> 00:26:06.225 post the question on LinkedIn, if that is fine. Yeah,
528 00:26:06.655 --> 00:26:07.655 Yeah. So I put
529 00:26:07.655 --> 00:26:10.065 a link on vis.io.
530 00:26:10.065 --> 00:26:13.065 Sorry, there's one more on vis io slash Discord.
531 00:26:13.325 --> 00:26:17.025 Um, that source another, uh, great, uh,
532 00:26:17.175 --> 00:26:19.585 channel community just like, uh, connect with us
533 00:26:19.645 --> 00:26:23.385 and other, other, uh, other developer who use,
534 00:26:23.925 --> 00:26:25.065 uh, Melva and Zoles.
535 00:26:27.045 --> 00:26:28.295 Okay. Thanks.
536 00:26:32.625 --> 00:26:35.035 Okay. Um, I think we're good for today.
537 00:26:35.255 --> 00:26:37.315 Um, really appreciate everyone's time.
538 00:26:37.575 --> 00:26:42.455 Uh, let's just, uh, um, keep in touch. Bye-bye.
Meet the Speaker
Join the session for live Q&A with the speaker
Ameek Singh
Solutions Architect, Zilliz
Ameek is a Solutions Architect at Zilliz where he helps enterprise customers takes best advantage of Zilliz’s best in class vector database solution. Previously, he was a Solutions Architect at Databricks and Sales Engineer at Fivetran. He graduated from UC Berkeley with a degree in Data Science