This week, we feature Arvind Nithrakashyap, co-founder and CTO of Rubrik, one of the world’s fastest-growing data management companies in the world.
Arvind Nithrakashyap has been in love with databases for a long time – two decades to be precise. Building a number of core Oracle database technologies, Arvind also co-founded the Exadata product line while at Oracle.
And while he loves his tech, Arvind not just a boring techie. His passion for travel has had him taking off to travel the world for six months.
For the most of his life, Arvind was pretty much like other geeks - busy building the next big technology, understanding customers and iterating products to suit their needs. What differentiates him, though, is his vision and the ability to build in a scalable and sustainable fashion.
But this was just a lead up – to something bigger and more beautiful – to Rubrik, one of the fastest growing data management companies in the world. This, and more, in today’s column.
Arvind’s father was born in a village in Mayiladuthurai in Tamil Nadu and moved to Hyderabad to study and later work. His mother was from Coimbatore and both his parents worked in banks. His father moved to Chennai and Arvind was born in Coimbatore in 1973. He was raised in Chennai and with both his parents at work, his grandmother was a big influence on him.
In 1979-80, Arvind’s father started his own business. “He was a huge risk taker, to an extent bigger than I am. He first got into the supply chain of gas cylinders, then he pivoted to the envelope business, which went quite far.”
Arvind says he is an introvert and as a child was never physically very active, preferring books over the sports field. A product of Sishya School in Chennai, Arvind recalls, “It was a very different school. Unlike other schools then, the emphasis on extra-curricular activities was very high with students from diverse backgrounds. It was like a startup of schools.”
Opting for Maths-Science was natural for Arvind, and he also took Brilliant Tutorials’ correspondence course to prepare for IIT-JEE. In his entire class of 30 students, he was the only one preparing for JEE. With an All-India 65 rank, Arvind joined IIT Madras.
IIT-Madras was a no-brainer and Arvind opted for Computer Science over Electrical Engineering, which was then a trend. His only exposure to computers was with Atari Games or ZX Spectrum Games and a few classes in basic programming at Computer Point.
Arvind decided to stay in the hostel, and this allowed him to interact with his college mates and forge many close friendships. “In the Computer Science department, we had eight PCs of which only three would work. We had Sun 360s workstations and not even SPARKstations.”
At IIT Madras, Arvind says he learnt more from his peers than his professors. He learnt PASCAL, C and Assembly Language programming in college and built a compiler, a GUI programme, and worked on distributed systems.
In his final year, his Computer Architecture professor Kalyana Krishnan was his advisor and was working on building local language support. While students from earlier batches had built a library that would take inputs based on phonetics, Arvind built an editor where one could type and save.
In his final year, Arvind got a job offer from Infosys but instead, he chose to study further an went to University of Massachusetts’ Amherst College, which was pretty big on artificial intelligence.
“During grad school I realised I wasn't exactly cut out for this - writing papers, going for conferences didn't interest me much. Two years into it, I decided to move out.” In grad school, Arvind worked in the software engineering lab understanding the synthesis between software engineering practices and databases. He worked on C, Ada and a bit of C++. Java was just beginning get popular then.
Arvind joined Oracle as a part of the university recruiting programme. “To me, it was very interesting because Oracle was working on the heart of the database system. I started working on some core projects around building support for 8K and 16K block size data warehouses. I wrote C code at Oracle for the next decade.”
Arvind was by far the junior-most engineer and his team was responsible for building a database cache. “Oracle allocates memory in database cache and when you query to read a block, you would actually pull the block, store it in the cache, pin it and try to keep the block as long as it remains hot. The algorithm would then track references to it every time there was a hit on the cache. I worked on optimising how we efficiently you write the data out, and manage LRU cache.”
One of his biggest learnings was that the things we take for granted – such as performance, and scale - are the toughest to build and achieve.
By 1998-1999, there were many things going on in Arvind’s head - whether to quit and start up or continue or get a green card.
His father then advised him to finish his work at hand at Oracle, go someplace else and start up after a year or so.
Looking back, Arvind says this advice saved him from being part of the dot-com bubble. Around the same time, a cousins convinced him to sell some of his Oracle stock, which crashed as well.
In 2000-01, huge multi-processor systems like 72 (core) CPUs by Sun, 64 (core) CPUs by HP were introduced. The team at Oracle saw that it was faster to run Oracle databases on 48 core CPUs than on 64 core CPUs (HP Superdome). This was because the database wasn't able to scale. The problem was in the recovery layer of the system.
“Database log was becoming the bottleneck, but nobody really thought about it in first place because recovery isn't really in picture. We rewrote the core recovery layer for Oracle and ended up multi-threading that part. It was a huge learning process for me - Go with your gut. People may say it doesn't make sense but you've to push through.”
Oracle had earlier worked with storage windows to get API access to better understand storage layers. The company thought it should own a storage layer and so Arvind and two colleagues decided to build the storage system. When he started, Arvind realised hardware procurement would take eight months. So, he simply went to the local store and bought PCs. The trio built Exadata for Oracle, which is now one of its core storage platforms. They used C++ to build Exadata.
It took us three years to launch the product version 1.0. By that time, a startup would have shipped maybe version 6 or 7.0. I decided that it was time to move on. I didn't want to continue to stay in Oracle just to figure out what I want to do later.
He quit and decided to travel with his wife.
Looking to work in a startup, Arvind joined Purisma, a startup in the customer management space. He took over the fuzzy match engine part, which worked on top of the Oracle database. Arvind worked on tuning database to extract more performance. This was the first time that Arvind had looked at database from the outside. Also, this was the first time that Arvind was working on the Java language.
Arvind worked with an external vendor to package the product and the idea was to make it very simple for a business user to install it. This was his first exposure to the sales side.
“I learnt that if you're open and honest with the customer, and you know what you're talking about, you can question their ideas and assumptions as well to make the project implementation successful.”
You may also like - Will over skill -- story of Vivek Vaidya and his $700 M startup exit
Purisma was acquired by Dun & Bradstreet and Arvind decided leave a year later. He wanted to travel, and saw this as an opportunity.
“We (Arvind and his wife) went to South America, starting from Argentina, spent two weeks learning Spanish, went to Bolivia, Machu Picchu trek, came back to Brazil, spent one month there. We then flew back to SFO for a couple of days and from there to London and then to Uzbekistan. Then we took the Silk Road. However, we couldn't go to Kyrgyztan due to visa issues. We went to Xinjiang, Kashgar, Beijing, Xian in China and spent two months there.”
I was jobless, homeless and probably in the happiest time of my life. When you travel, you see people of all age groups and different backgrounds. You see them falling in love with cultures and languages. I realised wherever we go in the world, there's Coca Cola and there's internet!
Back from his travels and Arvind met Bipul Sinha, the founder and CEO of Rubrik. The two had earlier worked together at Oracle. Bipul was an investor, and had invested in Nutanix. Arvind met Abhinav Gupta, founder of Rocket Fuel Inc, and decided to join the company.
It was interesting times as display ads were becoming big then. I built a real time infrastructure to be able to take requests in real time to show the right ads. When I left, we were handling 50 billion requests a day, and each request had to be handled within 50-60 milliseconds.
Arvind was the first engineer at Rocket Fuel and built a team of 30 engineers. He built the first implementation server based on the Google spec. The tech stack was MySQL database, and they had the Hadoop cluster where data pipeline and machine learning models would run. Every request went through the Hadoop cluster. Arvind and his team built an in-memory cache which would refresh at a fixed frequency. “Each of our servers acted independently. We built budgeting framework as well. We had data centres in Asia, Europe and the US.”
Related read - The untold story of Alan Cooper, the father of Visual Basic
Rocket Fuel went public in 2013, and Arvind was present for its opening on NASDAQ. By then, Bipul and Arvind had started talking about Rubrik. Bipul bounced a few ideas off of Arvind, and the duo started brain storming.
“Both of us believed there's something in this space. We got to know that almost a third of all backup customers would change their technology in the next three years, which means there was massive pain in the current systems.”
TSM, Veritas, Networker, and others were common back up options then. Arvind and Bipul saw demos of existing players and spoke to some experts, including John Thompson (Chairman, Microsoft). Everyone told them it was a tough space, but to Arvind and Bipul, this felt like the right solution.
Arvind believes they got very lucky with Rubrik and they haven't had to pivot. “The product we set out to build is the product we have today. The core thesis is same. You've a software that scales out, you put a policy framework in front, and then the system takes care of it (from infrastructure perspective).”
Rubrik – founded by Bipul Sinha, Arvind Jain, Arvind Nithrakashyap and Soham Mazumdar - raised $10 million from LightSpeed Ventures as the latter saw value in it. Arvind Jain was with Google and Soham had just sold his startup to Facebook. The three of them started coding right away, while Bipul played the role of product manager and recruiter.
The team set a milestone that in three months, they would have a proof of concept where they would be able to take a virtual machine, back it up on to a file system and then mount it back as virtual machine.
“We first built the meta data layer, I focused on that. Arvind (Jain) focused on file systems, Soham was initially looking at how to build the piece which allows to recover (data) instantly, along with data ingestion from VMWare.”
The team hired its first engineer in five to six weeks, and in two months there were eight engineers to work on job scheduler, core data lifecycle management, and many other layers. In 11 months, Rubrik already had its product deployed by a customer. They started the early access programme in 2015.
Rubrik was initially focused on VMWare and for almost two years, it was just supporting applications on VMWare. They started supporting physical SQL servers and Windows servers later and built a generic layer for data management.
Rubrik has been building distributed systems which could scale as needed. In the process, it built a lot of components and have to be careful how they design it. The focus is on simplicity, which solves a lot of problems.
“We have distributed task scheduler and a distributed cluster management system to detect error and auto heal. On top, we've a core policy engine which is data management layer (managed data lifecycle).”
There is a separate component building search index for everything, like search-based file recovery, and on top of it is the API layer. UI also interacts with this API. The team made this decision early on and it proved to be a big differentiator.
According to Arvind, Rubrik is at exciting stage with some interesting set of challenges (and opportunities):
Arvind says he doesn't make spontaneous decisions. “It has to be data driven. I take advice and feedback from others as well. I don't think you can make decisions in isolation.”
He believes the following line defines him well - “If I'm not good at what I'm doing, then I've no business being there.”
For now, he sees himself continuing to build Rubrik and taking it to its maximum potential. Apart from travel, he’s motivated to do something that has impact.
Technology to me is a tool that can have a huge impact on lives. What appeals to me is solving interesting problems and providing value as well.
Trivia about Arvind