Grey Beards on Systems - 44: Greybeards talk 3rd platform backup with Tarun Thakur, CEO Datos IO

Episode Date: May 2, 2017

Sponsored By: In this episode, we talk with a new vendor that’s created a new method to backup database information. Our guest for this podcast is Tarun Thakur, CEO of Datos IO. Datos IO was started in 2014 with the express purpose to provide a better way to back up and recover databases in the cloud. … Continue reading "44: Greybeards talk 3rd platform backup with Tarun Thakur, CEO Datos IO"

Hey everybody, Ray Lucchese here with Howard Marks here. Welcome to a sponsored episode of the Gray Birds on Storage monthly podcast, a show where we get Gray Birds Storage and System bloggers to talk with storage and system vendors to discuss upcoming products, technologies, and trends affecting the data center today. Hi, guys. Thank you very much. Really looking forward for this discussion. Thank you for the introduction there. Yes, my name is Tarun. I'm one of the co-founders and the CEO of Datos.io.

Briefly to just introduce myself, I started my career actually writing microcode, 32-bit assembly language code. Oh, God. We all know that stuff. Ray and I started with 8-bit code, but we're old. There you go. But people who write a semi-language code are the real people. That's what I tell everybody. Makes my brain hurt to think that small. You know, I used to debug that code back in 2002. And, you know, this was out in Minnesota, the Minnesota nice world. And it'll be snowing outside and you know this was out in minnesota the minnesota nice world and it'll be

snowing outside and you'll be sitting inside with a cup of black coffee and debugging hard-coded zeros and ones and trying to figure out what is going on so anyways you know i did started my career with with that at a company called seagate i was fortunate to have a team out of DEC Digital Equipment who came on board at Seagate to really help Seagate move on from disk drive-based business to more storage controller-based business. So, you know, some very famous people like Richie Lurie who built the early VAX PHP systems. From there on, I moved to the Bay Area,

met my co-founder, who's my co-founder now, Prasenjit, and he was like, look, you've done some interesting good work in self-healing storage. You should really be at IBM Research. And I was like, okay, if you say so. And I was lucky to be at IBM Research here in Almaden for San Jose for about three and a half years, then realized and really didn't want to be in the world of research after three and a half years and sort of moved into more product and market and sort of more product management role at Veritas side of Symantec, realized Veritas and Symantec is the worst marriage ever after a year and was, I think, very lucky and fortunate and blessed, I should say, to run into some very good people at Data Domain. And I was fortunate, again, to be part of that journey for four years, not in the

earlier years of Data Domain, but from late 2009, early 2010 onwards. I was fortunate enough to be with Brian and Hugo, the founders, for a while. And then, of course, Frank had built a phenomenal culture, phenomenal company, and a phenomenal sort of leadership team at Data Domain that really gave us everybody an opportunity to grow and really see how companies are built. So that's a little bit about myself. I come from a very strong product market background. The only one thing I enjoy is product management and now building a company. Okay, so tell us a little bit about Datos.io.

What's going on now? Perfect, perfect. No, thank you. So, you know, we started the company in June of 2014, you know, really sort of, again, going back to the life, I was at Data Domain, and my co-founder was at Life at IBM, sort of building shared nothing file system products. Both got together. You know, one of the observations we both had back in 2014 was, you know, look, really, the core element that drives the infrastructure choices is really what is called application. It's really the application that influences the choice of the IT stack, meaning a Oracle or a SQL server or a Teradata or a choice is really dependent on what application you are building on the top, right? Is it an IoT? Is it analytics? Is it transactional? Is it SaaS? Is it e-commerce? That influences the choice of database. Choice of database influences the choice of storage. Choice of storage has a strong influence on the choice

of data management tooling. Well, we like to think that you decide on storage first. But I recognize that you're probably right. Thank you, Howard. Thank you, Howard. Thank you. So, you know, that's essentially why we started the company. There are phenomenal transformative changes happening in the application tier, essentially two, one, the next generation cloud applications,

really the analytics, the IoT, the SaaS, are being born in the cloud itself on distributed architectures like non-relational databases. For example, the Cassandra, the Mongo, the DynamoDB, and your traditional applications are sort of migrating to the cloud as fast as they can. Maybe perhaps for non-operational use cases, maybe for DR, but all non-operational use cases of an application, traditional applications, are moving to the cloud at a very fast pace. And we can call that Platform 3 from IDC. We can call it Mode 2 from Gartner.

But really, given those two transformative changes, we think underlining, if the database architectures have evolved, if the application architectures have evolved, the observation we made is data protection is still sitting in a classical media server-based architectures. That's the problem we wanted to go tackle. That's how we started the company in 2014.

We are about two years, 11 months old. On the footsteps of our very, very big launch, which is TrueRoto in another couple of weeks on 2nd May. Okay. So the existing RecoverX product is backup for NoSQL databases. That's really interesting because my experience with database backup is that NetBackup and its ilk still treat databases like files. Like one big file. So my experience is that the NetBackups of the world treat databases like files. So my experience is that the net backups of the world treat databases like files.

And so you end up doing an incremental backup, which means backup the whole database because one record changed. How is RecoverX different? Perfect. No, thank you, Howard. I do want to take a step into zoom in into our 1.0 because it has been a transformative product. We got Gartner Awards. We got Product of the Year Awards. So I do want to spend a couple minutes on the product 1.0 because it has been a transformative product. We got Gartner Awards. We got Product of the Year Awards. So I do want to spend a couple minutes on the product 1.0.

You know, Luke Howard and Ray, really when we started, you look at the entire market of backup providers, right? The Veritas, Commvault, really the seminal, I would say the 80s, 90s era of database backups and file system backups, as Ray said. File system backups are one big file. As Ray said, file system backups are one big file. Let's keep backing it up every day, every hour. Then you got into this era with the rise of server virtualization in 2002. A company called Veeam came out from nowhere in 2006.

And look at today, 11 years fast forward, it's about a half a billion dollar company in running revenues, if not larger. And that's an interesting reversion to the late 70s, early 80s, back up the whole file system. There you go, with Legato and, you know, absolutely. And, you know, you're now sitting again 2010s and the cloud and sort of this whole application movement is such a rising transformative change now that, you know, there is no such thing as a LUN in a cloud. You go to Amazon and you ask them, hey guys, out of your $14 billion bookings, how much of that is ESX? And their answer is zero. There is no such thing as ESX VM in the cloud. You have to trust the data management has to fundamentally evolve from a LUN or a file system backup or a VM backup to an application

centric backup approach. Let's keep that in mind. Veritas to Veeam to Datos. We want to do to backup what Veritas did to database backups and what Veeam did to VM backups. We want to build the application centcentric backup company built from ground up for the multi-cloud world. Now on the NoSQL part, you look at that, that's a big vision, right? That vision is going to take 5, 8, 10 years. It doesn't happen overnight.

What we did is a cleverness or a strategic approach to the market, again, learning from Veritas and Veeam, is that you pick a market who there is nobody because backup is a very, very sticky market. That was the biggest learning I had from data domain. You know, we'd go into all the accounts and we'll try to replace net backup and we were never successful. It's a very, very sticky market. What we did is we figured out a piece of the market, which is called the non-relational database market the cassandra the mongo the couch the apache hdfs the apache hive and there is absolutely nobody there there is no veritas there's no comval there's no netapp there's no emc there's no legato there's

no vim and that gave us an opportunity to go into this white blue ocean greenfield market keep the company very very very well capitalized don Don't overblow. Don't keep going back to the well and keep raising money because we don't like to do that. That dilutes everybody. You go pick a market, keep it to a small 30% company and go get your first 20 customers and then grow from there. Well, how does somebody back up a non-relational database? And that's an excellent question, Ray. I think the real question is how do you back up a non-relational database. And that's an excellent question, Ray. I think the real question is how do you back up an eventually consistent database? There you go.

There you go. There you go. That's really the meat. Thank you, guys. So you're absolutely right. You take any of the non-relational databases, they're eventually consistent, right? They compromise consistency for availability, going back to the classical ACID principles or CAP theorem principles. So you give up consistency, you get availability. And that was fine for the first six, seven years.

And now, but if you look at, and these are public examples, I'm not quoting anything confidential. eBay runs all of eBay.com on MongoDB. You have HomeDepot.com, again, a public reference, runs their HomeDepot.com on Cassandra databases and Google Cloud. These are eventually consistent. They needed that because of high volume nature. Yet they need a backup and a consistent, a cluster consistent, a application consistent point in time backup. And guys, that's where we have about 18 patents.

We have reinvented both the notion of what does application consistency means in eventually consistent databases. And number two, big innovation pillar we have done is these databases run on shared nothing architectures. You don't run Cassandra on an EMC or a Nimble or a pure storage device. Then if you don't have a streaming block of data, you don't have a streaming block of data, you don't have a notion of a file system, how do you do deduplication for eventually consistent databases? And that's where we innovated this concept

of semantic deduplication. Let me just give you an example. What? Howard lives in Arizona. Arizona is where Howard lives. Semantically, they both mean the same thing. Howard lives in Arizona. Arizona is where Howard lives. Semantically, they both mean the same thing. Howard lives in New Mexico and resents being assigned to Arizona. There we go. So I'm sorry for that. Howard lives in New Mexico. New Mexico is where Howard lives. Semantically, they both say the same thing. Howard is the key. New Mexico is the value. Keys and values are

written in a very different way underlining when they're written in a database. So backup has always been important. I think that the budgets are set from CIO. Now the customers are putting pretty high mission-critical applications on these non-relational databases. And as you do that, the need for backup and recovery for these non-relational databases is becoming overly important. So does RecoverX actually understand the underlying schema of the data? There you go. We had to, Howard.

You had to understand. So we don't, we state that just like Oracle had our mind. Why don't you support Oracle? I want you for relational databases too. Yeah, I know. I'm going to come to that in a second. 2.0, it's all about 2.0. You're setting the T's up.

But let me answer your first question. We stay at a database APIs level, and just like Oracle had RMAN, Cassandra has very well-defined APIs to take change data capture. Mongo has very well-defined APIs for change data capture. We call them, but again, we invoke those APIs

to get the change data, but you do all the secret sauce on top of that. Let's say you, me, and Ray are three nodes of a database. How do you generate an application-consistent backup from three of our snapshots? And that's where our secret sauce is. I have to compare the schema and find the… Wow. I have to… There you go.

We run a... Thank you, Howard. You're almost there. You have to take the three snapshots, meaning you, me, and Ray, three nodes of a single database, and you have at a very high scale, you have to determine

what key values are consistent. Do you go so far as to monitor the timestamps on the ones that aren't consistent so you can find the last right? Dude, you guys are smart. Where have you guys been hiding? We try to be. We've been here for years. Yeah, but the snapshot's a storage environment version of the world, and the application has a different, you know, what about, I'm not a non-relational database expert, but logs and journals and all this other stuff.

So guys, we have cracked this nut. We have cracked this nut. We have about 20 paying customers now, people. We have, these are Fortune 100, Fortune 200, global 2000 customers running in cloud environments. So you guys are opening everything up and basically saving keys and values or rows and columns, not the files that hold the database. There you go. No files or blocks. That's exact because data is the king. In the cloud, there is no block. There is no file. Well, there's EBS, but that's all hidden. You never will find out exactly what is happening under EBS. So you're backing up the data. There you go.

Oh, that opens the door to all sorts of interesting things. I want to do two more podcasts with you, Ray Howard, because you are absolutely right. The data services, the data management services need to be at a data level, not at a file or a block level. And that's what the world needs to understand. This is a paradigm shift for my mind here. Well, this is actually,

this isn't even application aware. This is, you know. Data aware. Well, it's, you know even application aware. This is, you know. Data aware. Well, it's, you know, application understanding. So the database is sharded across these three nodes, Ray, Howard, and Tarun. And you take a snapshot of the data and then you go peruse the data to try to understand what's changed? Well, what's different?

What's different? What's different? Yeah. So because different. What's different. What's different. Yeah. So, so, so, and because there is no other mechanism, right?

You have, you have to go do application semantics. You have to understand the rows, the columns, the keys, and the values

to really determine what generates consistency. So you're generating a consistency point from an eventually consistent database. Now you're doing it post facto, but I don't care. Wow. So I'm actually tweeting to you guys right now because you have finally, Howard, you

finally have said things which we have been saying. You know, Curtis Preston, I'm sure you guys know, our friend, Mr. Backer, when he got the pitch, he was like, this is exactly how it is supposed to be. And yet you guys are not talking about it. So let's talk about what's coming up here, Tarun. You've gotten to the point of convincing us that the product works on these non-SQL databases and provides backup of the data in a consistent manner. Yep. So now let's go back to, you know, thank you, Ray. You know, we truly have built a pretty transformative product.

And so, you know, Ray, what was going on for the last one year? See, we launched this product in June of last year. So we've been at it for the last about 10 months now. And, you know, we've been very, again, fortunate to have won the confidence of all these Fortune customers. And they're like, look, guys, can you also now please help, back to Harvard question,

can you also please build this for SQL Server? Can you build this for Oracle? Can you build this for Postgres? And we're getting these questions about six months ago and we're like, why? Because you have Veritas, you have Commvault, you have Legato, you have Networker, you have all these products. Why do you want us to build? Those products back up the VM or the database. They don't back up the data. Thank you. And that's the answer we get, Howard. The answer we got from these guys is, look, guys, we got two pat answer. One, I don't want to care whether it's a physical server or a SQL server runs in a VM or SQL server runs in a hyperconverged box like a Nutanix.

I don't want to worry about a Nutanix backup or a VM backup. I want to worry about a data backup. Right. And so that was one part. The second part answer is, guys, I am moving this application to the cloud, perhaps for test F, right? Non sort of recovery use case.

And I don't want to take Veritas with me to the cloud. I don't want to take Commvault with me to the cloud. And we push them as like, guys, when you go from one home to the new home, you don't take your old couch with you. We don't want to do that. Well, I might take our old couches, but the rest of the world doesn't. I understand that.

Well, you don't take your old dishwasher. That's true. Dishwasher. That's a better example than the one that I used. And so, you know, that essentially led, that was the impetus to our journey for 2.0. We're announcing three key pillars of the announcement. Number one is all going to be about a use case called mobility. We call it the data mobility use case. Again, the example there being we are working with a very big customer of ours.

They want us to enable test dev of their SQL server in the cloud. So they want to keep the SQL server on-prem for various compliance and governance needs, but they want test dev in the cloud for their application developers, for their lines of businesses. So we're opening Cloud Mobility, underlining it, Howard,

going back to it's not a block or a file backup, underlining. We have now stretched this concept of semantic deduplication to the rows of a table within a SQL server. I don't do a block backup or a file backup of SQL server. Our deduplication is 10x of data domain deduplication. 10x of what data domain did. Because data domain at the end of the day operated at a boundary of blocks, 4k, 16K, 8K, because we had to worry about metadata blow up.

And there was a challenge of compression and all this other stuff that made it even worse. And you guys were smart in data domain and where you put through the block wasn't fixed. But if you know where the records and field borders are, well, then the block borders should be congruent with them. And that's going to boost your reduction rate, of course. Thank you. Thank you. So that's the biggest piece we have innovated. The other thing is that because you're taking a row rather than a block, it's a different orientation here, a different unit of change, really. Yes, sir. Yes. So the first piece is a wall-out cloud data mobility for relational databases.

This is huge for us. In 1.0, we cracked the nut of non-relational databases. With 2.0, we will be the only company in the backup and recovery space and the first company to crack the nut of relational and non-relational databases and provide mobility to the cloud. So is that like-to-like mobility? So MySQL to MySQL or SQL Server to SQL Server? Excellent question, Howard. Dude, you guys are smart. Omar, you guys are smart.

What I really want is to go from DB2 on-premises to Elastic Database. You guys are smart. You guys are smart. I'm telling you that. So we are getting that request. Actually, if you watch the Twitter feed of either mine or my co-founder, we got a request from a customer. Hey, guys, can

you help me move from MongoDB to MySQL or from MySQL to MongoDB? Can you do like to unlike also? We are absolutely getting that request. We're going to open up that use case future, confidentially speaking, more later half of this year. Right now, we're going to do like to like and then like to unlike later down the year. Okay. So that's the first piece, Ray, mobility. Number two piece is going to be around, you know, we

already do non-relational databases. There's, you know, file system backup use case that we talked about earlier that is very much growing as an important use case of Hadoop. Sort of the digital repositories, right? These repositories are now becoming half a petabyte, a petabyte. You know, the use cases of archival and tiering and even backup for subset of your directories of a large file system like Hadoop is becoming very critical. So we are opening the use case of data protection for big data file systems like HDFS. We're starting with Cloudera and Hortonworks. And we're working with Oracle also because Oracle big data uses Cloudera internally. So that impetus was really from the team at Oracle.

That guys look, 90% of our customers are financials. They need backup and recovery no matter what data storage you need. And they are right now have a ton of data on these data lake repositories. And so that led to us building the backup and recovery for HDFS. File system backup use case is always amazing. And the third piece of our 2.0 launch

is all about our underlining platform. You have to keep investing in underlining product itself. So we are doing three things there. Number one, Datos can be deployed as three-node software today. We're extending that from three to five nodes. Again, going back to since we are finding the consistency across keys and values, you need a lot of compute horsepower. Yeah, you're not just a data mover. You're interpreting the data in flight. So you're going to need more CPU. You need compute.

There you go. But that CPU complex is scale out. That's exactly right. So you can scale it out and you can scale it back. So it's truly elastic. So we will let you go from three nodes to five nodes and five nodes to three nodes. That's one.

Number two. Can I run it in a container? There you go. Did you see my two-door deck? Did Peter send that to you? No, no, but I run backups for a specific period of time while the backup is running. And so I want the scheduler to spin up the containers

and do what you're doing and then throw them away because I want to use that compute for something else. Clearly now I know why you guys have so many lovers around the world. So Howard, you're absolutely right. The more sort of a customer-centric feedback to us was, look, guys, most of these deployments are in microservices environments, right? Either a Docker environment or Mesosphere or cloud with AMIs. Can you give us, don't give me a tar.gz file of RecoverX. Just give me a Docker image or just give me RecoverX as an AMI.

Yeah, worst case, I want an AMI or an OVF, and I want you to be scale out, and only one of the VMs runs 24 hours a day, and it spins up the other ones when it needs them. Yeah. But ultimately, you're doing something that fits containers just so well. Yeah, so we are announcing the second part that we are announcing is Datos RecoverX

will now be available both as AMIs or as a Docker image, again, to help in these microservices environments. And the last part of our underlining platform is around policy management. Guys, you know, policy for backup and recovery admin is really their bread and butter, right? I set up backup policies in this little bit of much more

dynamic environment. You start with a policy that I want to protect this table, I want to protect it at this frequency, at this retention period. And they get new tables. At the table level? At a table level. At a table level. They don't want to do database level backups anymore. People want table level backups At a table level. At a table level. They don't want to do database level backups anymore. People want

table level backups. And they're like, hey, we got three more tables from these three application developers. I don't want to define a new backup policy. Can you automatically, on the fly, add those tables to an existing backup policy? So we added a ton of policy management capability,

the ability to resume and suspend, suspend and resume, to be able to add tables on the ability to resume and suspend, suspend and resume to be able to add tables on the fly to an existing backup policy. So we added a lot of that capability. Can I set a default policy per database? So any new table in this database gets this level of protection? Yes, yes, you could do that now with 2.0.

You can have default policies configured with a default retention period and a default frequency. Sorry, version backup interval. Yeah. Okay. You talked about the first piece was cloud data mobilities, and we talked about this like to like and like to unlike. So you could actually take a, I don't call it,

a SQL server database that's running on a host environment and provide the migration to the cloud of that? That is correct. We are in 2.0, Ray, we're going to announce, if you have a SQL server running on-prem, regardless of it is running on a VM physical server or an HCI, it doesn't really matter to us because we are not bound to that. We will give you the capability through RecoverX to migrate that data,

that application to the cloud, and you can migrate only a table if you want, not the entire database. Oh, God. And insert that table into a SQL Server AMI running in the Amazon cloud. The world of backup and recovery, it's so exciting. It is, it is. Well, it's more than backup and recovery. Yeah. Just because you're so fine-grained i mean the you know curtis and i both rant pretty regularly about how backups and archives are

different but you know there's an archive use case come version 4.0, that starts getting really interesting too. Yeah, yeah. And, you know, just to finish on that 2.0 preview, in addition to the product, we are announcing a ton of go-to-market stuff around partners. For example, we are working very, very closely with an app team. You know, they look at our install base and they're like, guys, excuse me, 80% of your accounts are our accounts.

How come you're going into our accounts and selling backup for all these applications and they're not being stored on NetApp? Yeah. So we're announcing them. Speaking of which, so the back end is a file system or an object store? There you go. We can either store to an S3 like Amazon or an object storage, or it could be an NFS of NetApp or of Data Domain. It doesn't really matter.

No tape? No tapes yet. Remember, he worked for Frank. No tape yet. No tape yet. People are asking, can you rather one year on NFS and next six years on Glacier? That's what they're asking for. Yeah.

Well, there's the object storage as the backup repository is really becoming a much more interesting concept. I think we're running short on time. Yeah, yeah, yeah. So is there anything you'd like to say, Tarun, to the listening audience as a final discussion here? I don't have any further questions, Howard. That would be the other question. Howard, do you have any final questions? I mean, I have a million half-formed questions. Yes, yes, I know.

And really want to get my hands on 2.0 because since I don't run NoSQL databases myself, 1.0 wasn't terribly interesting as a I want my hands on it. But, boy, this is an enormous paradigm shift and just the way we should have been dealing with databases for a long time. Howard, I am going to quote you on that. I need that statement in quotes. You know where to find me. quote you on that. I need that statement in quotes. You know where to find me. I already tweeted that. So I,

but no, it's Ray and Howard again. Thank you so much for the opportunity. We, you know, we've been pretty hard at work for the last three years, really operating. Oh, it shows.

Yes. Thank you. Thank you. I've really been operating the company as a submarine. Don't, don't like to make too much noises. Just stay focused on building and executing. And really, I couldn't be more proud with what we have been achieved with 1.0.

We're going to see a tremendous success with that product. And 2.0 is going to open a significant, significant amount of time for the company. And also help us address the journey with the customers that are on the cloud. Right. It'll help us be part of that. And so that's the, you know, mobility is going to be huge for us. It addresses the customer use cases.

It helps us be part of. Oh, and it ties directly into all sorts of things. I mean, I'm doing a workshop at Interop next month on managing storage on-premises and in the cloud. And the whole data fabric. We'll be talking next week. NetApp's data fabric is pretty big in this space as well. So, yeah, you're absolutely right.

Yeah. Yeah, just hold on to the NetApp data fabric. You will, you know, we're working very, very closely with that team on how Data Fabric. And so hopefully we'll have some good stuff to share. All right. Very good. Well, this has been great, Tarun. Thanks for sponsoring our podcast today.

It was a pleasure. It was a pleasure, guys. Ray, thank you very much. On our next monthly podcast, we'll talk to another startup storage technology person. Any questions you want us to ask, please let us know. That's it for now. Bye, Howard. Bye, Ray. Until next time.