Raj Yavatkar, CTO, Juniper Networks

The Equinix Vision for Self-Driving Networks for Interconnection & the Edge | Juniper Global Summit

Global Summit 2021 Leadership Voices Data Center
Raj Yavatkar Headshot
Raj Yavatkar, CTO, Juniper Networks, seated in a chair,  looking left while talking in a video to Muhammad Durrani of Equinix about brownfield deployments, intent-based networking, and network as a service.

Automation and the Self-Driving Network

Described as the world’s digital infrastructure company, Equinix has built one of the most trusted, reliable, and secure networks in the world. In this interview, Muhammad Durrani shares how Equinix is leading the way to a Self-Driving Network for interconnect and the edge.

Show more

You’ll learn

  • How Equinix utilizes automation, network programmability, and AI

  • How an all-inclusive data lake drives insight and action

  • How Equinix provides an exceptional experience for its customers

Who is this for?

Business Leaders Network Professionals


Raj Yavatkar Headshot
Raj Yavatkar
CTO, Juniper Networks

Guest speakers

Muhammad Durrani
Sr. Director, Global Network Architecture, Equinix


00:00 [music]

00:07 -Hello, everyone.

00:09 My name is Raj Yavatkar.

00:10 I'm Chief Technology Officer at Juniper Networks.

00:13 It's my pleasure today to welcome Muhammad Md. Durrani,

00:17 who is the Senior Director of Global Network Architecture at Equinix.

00:22 Md., welcome.

00:23 -Thank you, Raj.

00:24 Pleasure to be here.

00:25 -Your last name really brings some fond memories to me because

00:28 I grew up in India and I was a very big fan of Indian cricket.

00:31 My favorite player growing up was Salim Durani,

00:34 who was a flamboyant batsman.

00:37 Welcome.

00:38 Thank you. I appreciate it.

00:39 -Juniper and Equinix have partnered together for many years now.

00:43 We're very happy to have you as a customer.

00:46 What I find most impressive is that you've built one of the most

00:50 trusted, reliable, and secure network which is worldwide.

00:54 Can you say a little bit more about how you have done it?

00:57 -When customer comes and buy colo space from Equinix,

01:00 they need an interconnection portfolio where

01:04 they can connect their data centers to their branch offices.

01:07 They need trusted internet so they can browse an internet.

01:10 They need connectivity with the Cloud providers.

01:13 For that, Equinix needs to build something that give very good

01:17 experience for the customer of its single point click they can have

01:20 that experience of connecting their data centers throughout the world.

01:23 This is where we have to partner with Juniper and we have to buy

01:27 some carrier-grade routers where we can implement technologies

01:31 like segment routing, MPLS Traffic Engineering, VXLANs worldwide.

01:37 Then all that need is programmability capability to an orchestration,

01:44 give customer an API experience to build

01:46 that digital ecosystem that they want to build.

01:48 -That's good. Thank you.

01:49 I always wonder that you have more than

01:52 200 data centers all over the world.

01:55 You have interconnectivity among them.

01:57 You have wide area network connectivity for your customers.

02:00 You provide interconnection to their branch offices.

02:03 You're basically operating in multiple networking domains.

02:07 When there's a problem, how do you pinpoint a particular domain

02:10 and problem within that domain?

02:12 -Totally.

02:12 I think a very good question, Raj.

02:14 I think this is-- We're evolving in how to make that self-driven network.

02:20 It is a continuous effort.

02:21 There are jargons that you're listening about AI and ML and all.

02:25 To build a self-driving network, you need to have rich database.

02:30 You need to grab as much information possible from your network,

02:34 so you can build a correlation when the event happens,

02:37 how you correlate and pinpoint a problem.

02:40 There's a brownfield networks.

02:42 There's a greenfield networks.

02:43 Brownfield, you don't have technologies like streaming telemetry.

02:47 Brownfield, you have SNMP to deal with.

02:49 Greenfield, you can implement streaming telemetry.

02:53 For example, let's say, customer wants to have

02:55 information on how much traffic flowing through your network,

02:58 coming in from untrusted domain to me, and all.

03:02 For all of the work, you need to build a very sophisticated automated

03:06 system with very rich data sets so you can run all of these correlations.

03:11 Equinix is continuously evolving.

03:13 We're not there, but we're getting better.

03:16 Thanks to our partners that they're building features for us

03:19 to make us enable to do and go towards this self-driven network.

03:24 -That's very good, Md.

03:25 I understand, but I always wondered about the following thing that as

03:30 we are here to help you, we are also making a big pivot decision.

03:35 You notice that we call our NorthStar experience-first networking.

03:40 Which means it's not enough to provide high performance, high quality,

03:44 reliable network, but we also want to make sure that our customers, you,

03:49 have the simplest operational experience using our products.

03:52 For example, we introduced Paragon Automation Suite,

03:56 which is for our WAN products to really make it easier to operate them.

04:00 We acquired a company called Apstra Networks,

04:02 which has intent-based networking and closed loop automation

04:05 to make it very easier to manage data center fabrics.

04:08 As we make this, we would then start delivering experience-first.

04:12 I already know that you have lots of large customers, very happy customers.

04:17 The question is what do you do to make sure

04:19 they have the best user experience,

04:21 and what can we do to make it easier for you to do that?

04:24 -Very good question, Raj.

04:25 What we expect from our partners, vendors like you is to give us more

04:29 programmability on the network there.

04:31 The more API you can give us, more we can automate.

04:36 More we can automate, more we can make them more self-driven,

04:40 and we can give more control to the customer.

04:43 They can own their own ports, own part of the network.

04:47 Through an API, they can automate it.

04:50 Intent-based is what the next-gen networks are.

04:55 Customer can come-- On your orchestration portal,

04:57 they can come and define their intent of application.

05:01 Based on that intent, we can carve out network into different intents,

05:06 for example, low-latency intent, high bandwidth intent, best effort intent.

05:10 Based on the application SLA that customer provides,

05:14 we can give an AP experience to the customer

05:17 that with the single click or single API,

05:20 they can provision their data centers.

05:23 They can have their traffic picked or sliced in the network really

05:29 meeting their intent of the traffic.

05:31 Great that you guys acquired Apstra.

05:34 What we're looking from our vendors is to give us more control

05:38 on the programmable level

05:39 rather than CLI because automation using CLI is pain.

05:45 I think it is the right direction that we're going towards.

05:48 -Thank you for that.

05:50 I think one of the things we're also doing with experience-first

05:53 is use of telemetry and analytics.

05:56 I wanted to understand from you what kind of telemetry you already used,

06:00 and how do you apply big data analytics to that, and what can we

06:04 learn from you so that we can start building that into our products?

06:08 -I think most of the vendors, what they're doing is they're giving

06:12 us transport and an encoding.

06:15 For a company like Cloud,

06:17 like Google of the world and Microsoft of the world,

06:20 they have their development armies behind.

06:24 They can build a data lake who can suck up the data from

06:28 a different type of hardware.

06:29 Service providers are not that sophisticated.

06:33 What we need from vendors,

06:35 since the automation of the platform is done,

06:37 is to invest mostly on this data lake, multi-vendor,

06:42 multi plugin type of data lakes,

06:45 where we can get data from SNMP step of the world.

06:48 We can get data from streaming telemetry.

06:50 We can get data from S flow, J flow, IP fix.

06:53 There's so many variations.

06:55 If you go multi-vendor, you have three vendors, your host.

06:57 You have to build that, and it has to be very sophisticated.

07:00 These telcos providers, they do not have this sophisticated

07:05 network development experience.

07:07 If vendor can help us build that data lake as well, and then there

07:11 are algorithms available for MI, ML, and AI, we can run those

07:15 algorithms on top of the rich data set that we grab from the devices.

07:19 Then what we're driving toward is mostly intent-based self-driven network.

07:24 Intent is one thing, but then when a fault happens,

07:27 how do we figure out there's a fault?

07:29 How do we self-correct it?

07:31 At the time I'm self-correcting the problem,

07:33 how I can see the customer data towards the working links

07:38 to give them more resilient, highly available all the time type of network?

07:44 Next, I think, drive from vendors should be to go further ahead in

07:49 the stack and invest in building that comprehensive data lake with

07:54 multi plug-in, multi-vendor, and multi-standard transport encoding.

07:58 -Got it.

07:58 Thank you.

07:59 Md., one of the things I noticed recently that you are taking a big

08:03 step by is trying to offer networking as a service to your customers.

08:07 That's a big development.

08:08 Can you say a little bit more about what you plan to do with that?

08:11 -Yes.

08:11 Equinix has two sets of broad products.

08:14 One is interconnection, which is

08:16 pretty much comprises of physical networking devices,

08:20 Juniper, and other vendors.

08:21 Second set of products that we're building are mostly the edge services.

08:24 We want to build a global edge platform

08:27 that can offer customer a bare metal as a service,

08:30 and then customer can also bring their virtual network functions.

08:34 Now, you have both sides of the world.

08:36 Then if you are in a very small data center,

08:40 we have capability to offer exactly same

08:43 feature-rich interconnection portfolio for our customers.

08:46 -the smaller footprint.

08:47 -Totally.

08:48 Smaller footprint, low-cost entry, and then give customer a ways

08:52 to connect their branch offices through this virtual infrastructure

08:56 to back-office in a regional data centers or a colo data centers.

09:00 This network is a service you want to abstract it as customer has no

09:03 clue what's underneath, whether it is a physical or it is virtual.

09:08 Look and feel should be exactly the same.

09:09 Experience should be exactly the same.

09:11 For both customer at Equinix is low cost as well

09:15 and depends where you're getting the service, and high cost as well

09:18 depends if it's physical versus virtual.

09:21 From a user experience point of view,

09:23 more experience is API driven

09:25 and give customer easy ways to connect anywhere,

09:30 any place within an Equinix data center.

09:34 -Thank you.

09:35 I have a last question, really in two parts.

09:38 I hope you address the both of them.

09:40 One is that, what is next for the networking industry, according to you,

09:45 and more importantly, what is next for Equinix?

09:47 -Totally.

09:48 Next for Equinix and for industry is--

09:52 I think security is a critical function.

09:55 When we start building our edge platform--

09:59 We can do networking very easily.

10:01 It's in our DNS, but when it comes to security,

10:04 it becomes extremely complicated.

10:06 There are two approaches to a networking you can offer

10:09 in a virtual infrastructure.

10:11 You can have VNFs and all these, all centralized.

10:14 Centralized data approach doesn't scale well because if you have colo

10:19 customers-- The challenge that Equinix has specifically-- I'll talk about it.

10:22 Colo customer has 10, 20, 100 gigs of links come in and forwarding

10:26 traffic within Equinix data center.

10:28 If you want to provide security for the customer and you go

10:32 with the virtual VNF, these VNFs has fixed bandwidth of 10 gigs.

10:37 Now, 100 gig traffic coming hitting my 10 gig right there

10:40 am I hitting a bottleneck.

10:41 If you go a physical appliance, the same thing.

10:44 They're so expensive.

10:45 These physical appliance from security vendors that getting

10:49 a bandwidth with security is extremely expensive experience.

10:53 -When you say security, you mean firewalls as well as packet

10:55 inspection, all that stuff?

10:56 -Totally.

10:57 IPS/IDS and stateful firewalls.

10:59 If we start offering those functions inline.

11:03 I have racks, X number of servers, and it is top of the rack,

11:08 none of the vendors so far is building the inline function.

11:12 We are not asking full-fledged firewalls.

11:15 Very few functions in the firewall, IPS/IDS is one of them,

11:19 but then stateful security groups, L4 and L7 load balancers, and routing.

11:25 If we just build these three functions-- Scale is not very huge because

11:28 you're not talking about having 1,000 customers on their top of the rack.

11:33 Typical configuration is 32, 42 servers in a rack.

11:37 If every tenant get one server, you have maximum 42 tenants.

11:41 This is rough math.

11:42 You can provide these functions inline and distributed manner.

11:46 That's the industry game-changer.

11:48 None of the vendors are doing it.

11:50 We're talking to a few vendors including Juniper to build something

11:54 that can give us huge scale for this edge platform data center.

11:58 Security, I feel is a huge progress if you can make in inline manner.

12:05 Second is, as I told you, building the self-driven network.

12:09 It is NorthStar.

12:10 You can go step by step.

12:12 As I told you, the challenges are brownfield and greenfield.

12:15 It is just not greenfield.

12:17 Greenfield, it's pretty easy to build these--

12:20 -Everything is new.

12:21 -Exactly right.

12:22 I have 80% of my network brownfield.

12:26 Now, I have legacy tools, legacy SNMP with new streaming telemetry.

12:31 Then if you have to sample the traffic, there are S flow, J flow.

12:35 Every vendor has their own flavor of software.

12:37 Building that data lake with all of these technologies as a plugin,

12:42 and then running this ML and AI algorithm,

12:45 which based on the rich data set will give you a correct information

12:49 and more accurate self-driving approach to the networks.

12:52 As I talked previously as well, getting an application SLA and

12:56 implementing it and taking customer traffic, putting it into right slice

13:00 in the network is an easy job, but what happened when failure happened?

13:03 This is where the self-driven network,

13:05 finding what's the problem your network,

13:08 how do we see the traffic to the next best available link by also meeting

13:12 the customer SLA is a critical thing.

13:14 I feel like much towards the self-driven network, which involves building

13:18 these data lakes, multi-vendor, multi-protocol, multi encoding,

13:22 multi plugin, and then inline security firewalls, IPS/IDS load balancers

13:29 will be a game-changer for data centers.

13:32 -Md., Thank you Very much for your time today.

13:35 I really appreciate your insights.

13:38 Some of the takeaways I took from these

13:40 conversations are very important.

13:41 First of all, we must provide APIs.

13:44 We should have support for open data lake

13:47 that can support telemetry coming in from multiple vendors.

13:51 We should have inline security, layer 4, layer 7,

13:54 even at the top of the rack switches

13:56 so that each customer can be supported separately.

13:59 Finally, the vision of self-driving networks

14:02 should also be applied to brownfield networks,

14:05 not just greenfield networks.

14:07 With that, thank you again.

14:09 [music]

Show more