High-Performance Storage – Kam Eshghi, Lightbits Labs

This is texturing TV. Hey guys. Thanks for the throw.

We're here with Kim SG. Who's chief strategy officer for light bid labs. They just picked up 42 million dollars in funding the drive the next wave of cloud storage can't welcome to the shop.

Thank you very much. We've had cloud storage for a while. And of course there are new technologies coming down the pike.

But why do we need a new platform and what's different about what you guys are doing? Yeah, so what we provide is a high-performance block storage service that can be used in Cloud infrastructure that can be either internal clouds in an on-prem environment or with cloud service providers, whether they're big public Cloud providers or sort of the edge providers and the idea here is to to make high performance disaggregated storage simple agile and efficient. If you think about the different ways storage is being used in in the cloud today you You have Solutions out there that are disaggregated and our software-defined, but they don't quite meet the performance requirements of today's native Cloud database applications on the other hand.

If customers end up using direct attack storage to be able to get to high performance they end up wasting a lot of storage capacity because you don't have high utilization of your storage infrastructure in that way. What we've done is we've come up with the model where you can have the segregated storage which means pooling all your storage together and in a way that can be shared across all the application servers and in the cloud and at the same time give performance that's equivalent to local flash so we can be used for high performance database and analytic workloads that are latency sensitive and at the same time give much better utilization of the infrastructure which means lower costs and data services. The very long answer to your question, but it is a you know, our solution was designed from the beginning for the cloud model, but for applications that need very high performance.

Is it your sense that more of those applications would be moving into the cloud now because of this kind of technology is available. I mean, we've had block storage forever in the databases and run on premise and largely for the reasons you point out latency, but is nvme and block storage gonna make it easier to move those applications in the cloud or you think they'll still run primarily. It's Now I I think they are moving to the cloud and you know what we provide if you go back to how life it started now, we we that we want to focus on cloud designated storage and initially our Focus was on private clouds and Edge clouds.

And the first thing we did was we invented a new protocol called nvme over TCP that allows nvme to provide disaggregated storage on top of existing ethernet cpip Network infrastructure. So that was a big big step forward just to be able to make it as easy and efficient as possible to provide the segregated storage on existing infrastructure. The cloud is using, you know, not RDMA infrastructure or fiber channel, but just ethernet tcpip vanilla ethernet CPAP, and after we created this new protocol and work with the help of Intel and Facebook and others, we got it standardized then we came out with the Production solution around this and added data services added resiliency and high availability to the solution.

So now it's a complete storage solution. It has been for the last few years that that is deployed across, you know on-prem as well as cloud service provider infrastructure. Now that this is available.

And by the way nvme over TCP is part of the Indian River Fabrics standard. It's basically another transport option now that it is available. It makes it much more efficient to to run these high performance database applications in the cloud today.

If you look at what's available high performance cloud storage is expensive. And it's complicated and you have many different options that you need to look at and then depending on the workload requirements you pick the right option. What ends up happening is if you go with the highest performance option with other solutions that are out there you you pay a lot for that service whereas with light bits you can get to much better TC or much better price performance for those high performance applications and our customers can get the same consistent experience whether they run like it's on Trend or in the age or in the public Cloud now our public Cloud offering is not out yet, but it is about to come out we are doing a release which is for early customers next month.

And so we'll keep you posted on updates on that not something we're announcing today, but it is it is about to be released and that would be a light that's on AWS offering and we'll follow with other public Cloud off. with the big Piper scalars Do you think as time goes on we'll see more these applications running in a hybrid mode that is result. You know what I have a more distributed or do you think that because of latency issues that either run in the cloud or they'll run on premise but they won't be as Loosely couple.

How far can we get? That's great question. You know, we see a lot of different models.

I think for sure for the big Enterprises that have a certain amount of data, you know, once you reach a certain critical mass, there are economic advantages to having your own internal Data Center and when I say internal data center, it can still be a cloud model on Prem where You are doing software-defined everything your infrastructure can support multiple applications. It's very agile and flexible the same thing that the big cloud providers are doing but it's done internally for the internal customers that that Enterprise has so there are economic reasons for continuing with on-prem Cloud infrastructure, but then those same Enterprises would like to be able to expand their infrastructure rapidly and and use the public Cloud for that or in certain geographies. They don't have data center.

They want to use a public Cloud for that. So they they'd like to get the same experience on-prem and in the public cloud and that is one hybrid model where their data resides in both environments and their processing is also done in both environments, but they're looking for a unified and consistent experience. Then then you have some other customers that are born in the cloud and they're entirely in the crowd Cloud.

So they're not quite hybrid. They may have small data centered internal development deployments, but it's mostly in the cloud. and and we're seeing some other examples of edge Cloud providers or some tier two csps cloud service providers that actually provide infrastructure for the storage that then works with a multi-cloud environment and and they essentially are moving any sort of lock-in that one of the big public Cloud vendors would have by Expanding to a multi-cloud environment.

So there's many different many different options out there. The way light that started is our Focus was initially on private clouds and Edge clouds and and now we're having light bits available on AWS and soon after that on Azure. So our goal is to make high performance cloud data storage very simple and efficient and at the same time allow our customers to get the same type of experience and high performance block storage service on either of these deployment options.

The other thing that we've done is we have spent a lot of time building a ecosystem around nvme over TCP and developing our solution alongside various different application environments that are commonly used for both containerized applications with kubernetes virtualized applications VMware also with openstack so that An application point of view it's very easy to basically run on top of the libitz cloud data platform and get this high performance block storage service with the user experience that that's very familiar to them from a management and orchestration perspective. Will I need your traditional storage administrator in this new era or is the cloud essentially just another resource that I'm invoking through apis or whatever and compute and storage is all converging and we're just managing this in one kind of holistic model. Yeah, it is definitely moving in the direction of more of a holistic model.

This is not the old you know, it infrastructure Legacy it infrastructure that required social specialized specialized people for storage versus compute and networking. There is some separation still in some of the larger organizations in terms of the Specialties that infrastructure Engineers have at our customers, but but the idea is to make this as easy as possible to be able to deploy storage almost like an application and and kind of attached to it and use it, you know, if you think about how we get deployed it's it's very simple. We we run on commodity servers.

So this is not we're not providing any custom Hardware Appliances. We can run on Intel AMD and soon armbase servers and and those servers are software pools all of the storage across. The cluster of servers and can application servers that are anywhere in the in the data center can attach to volumes within the storage cluster and get high performance block storage.

So it's it's easy to deploy. It's easy to scale as the customer wants to scale the infrastructure and you get at the same time low latency and all the data services, you would expect like compression and thin provisioning and as well as high availability with with raid and replication. So the way we get deployed the way we get managed the way we can scale is very very simple and the idea is to have a holistic view of the entire Cloud both on the application side and on the storage side and working on top of the existing Network infrastructure.

Can you think that people are fully appreciated the profound I would rbu implications of nvme and what that means to folks or you know, is this one of those Technologies where everybody kind of takes advantage of it without realizing exactly how much it changed the way we think about storage? That's a good question, you know nvme as a technology has been around for a long time now and any new technology like this or protocol like nvme takes time to get adopted but the whole idea with nvme was to create a host control interface for storage that is designed for Flash not designed for hard drive, which was what was used in the past with data and SAS. So the initial introduction of nvme was for direct attached ssds it it was designed for Flash.

So it really made the most of the flash a performance and latency that that's available and and it took time for it to be really adopted and and become the dominant interface for for ssds and it's really only been a few years that nvme SSD became the most popular form of SST, especially for Enterprise and data centers. then the next major Improvement there was to to think about how can we put all this high performance flash on the network and not have to only use direct attached storage, but be able to disaggregate storage from compute and get all the efficiencies that come with that the flexibility to scale your storage and compute independently and the ability to Make, you know have much lower TCO infrastructure by having higher utilization of the infrastructure and and also providing centralized data services and high availability for that storage. So moving nvme flash onto the network was was how nvme over Fabrics was born initially with RDMA fabrics and then the third major breakthrough here was the work that light bits did with our partners both the introduction of nvme over TCP, which really makes nvme over fabric suitable for the cloud because now you can run it run this protocol and give me a Fabrics on top of existing Network infrastructure of vanilla, ethernet CPAP and and be able to still get high performance low latency and segregated storage.

So to answer your question. Absolutely I think nvme is is really allowed us to make the most of what's available with the Flash. To express the full value of The Flash to the application and with nvme over TCP and then all the work that we've done not just to introduce nvme over TCP but clustering for full scale-out ha solution intelligent flash management to to reduce State latencies and and provide data services, you know together with nvme gives you a complete storage solution.

Do you think also the way that we build and run applications will be different in this regard? Will I see more software components may be running closer to the storage rather than the compute because I want to run analytics directly on the Block but maybe the computer the rest of the application is running on a traditional server somewhere Mr. Way.

We need to think about the way applications are built into Point changing. Yeah, well today we're saying applications being developed and deployed in you know, some of the common environments such as VMware and and kubernetes and openstack and they are still disaggregated from the storage side. So the storage is providing high performance low latency volumes, you know with light bits and the application is sitting on application servers that are elsewhere in the infrastructure and multiple tenants and multiple heterogeneous application environments can share the same storage cluster.

Now we have every once in a while, you know we hear about and we talked to some customers about the potential value of doing more computation on the storage itself it particularly for some use cases such as for AI and machine learning and training there are some advantages to doing that and certainly in the light that's architecture it you know, we're well suited to take advantage of that because you know, you have all the storage across a number of storage servers that also have, you know, competitional horsepower that can be used for some of that computational storage model as of today. You know, we're not seeing a large Market opportunity there, but certainly over time that could change and I'm specifically talking about the use cases of life. It's just focused on which are again low latency database applications and also performance sensitive analytic workloads that I would like a more general purpose high performance block storage service.

All right, cool. Well, you heard it here first guys storage is cool again. Just a question how we're going to use it cam.

Thanks for being on the show. Thanks for having me was pleasure. All right back to you guys in the studio.