A: Rackable acquired SGI about a year and a half back and changed the name of the merged company to SGI. What has changed is the addition of the cloud customer base in addition to the HPC (high-performance computing) and visualization customer base.  For me as a CTO, who has been with SGI for more than 20 years, i have seen the addition of this new customer base on the cloud side. I see a lot of leverage on both sides – cloud and HPC visualization customer base.

We are a major supplier to cloud companies such as Amazon. We have shipped hundreds of racks to the Amazon cloud. When you touch the Amazon cloud you are actually going through an SGI system. if you are using a motorola Droid, you are using an SGI system in the background. Increasingly, we have morphed into a server-and-storage company.

We have now gone in the backend and have diversified. But, we still have the know how of the data work flow of the movie industry, the cloud and HPC. We have now moved one step back into the server and storage side and at the same time diversified to embrace and accommodate the front-end.

A: We always push the envelope and that’s how we differentiate. on the movie industry side, we differentiated ourselves by ensuring the movie frames in our servers move in an uninterrupted way and we can give the reassurance that not a single frame is dropped. for the movie industry, each and every frame is intellectual property. on that end, the i/o and data throughput is taken care of by us. Then on the HPC side, we are trying to move from teraflop, to petaflop to exxaflop. The problem has been space and power. We are working on the power side right now as a long-term effort. But, for space, we are always working on packing more compute power in a single cabinet. So, Peta-fop-in-a-box is a first step to that goal. We do this by using graphics-processing-units or GPUs in a small footprint. But, this is not free lunch – there is a lot of re-programming required to use GPUs instead of CPUs to get more fops in a box. if someone is ready to do that re-programming we are ready to give them peak-peta-flop-in-a-box today.

A: If you take different levels of computing, the traditional level is one CPU core and one piece of memory. When you are writing code it is running on this single core and single piece of memory. The next level is when you need more power from one CPU, you are linking multiple CPUs together under one operating system. Here you want your program to split into various parts to use all the available CPUs. This level of reprogramming is not extreme as you use the open MP model which allows you to use more CPUs. The next level up becomes more difficult where the CPUs are not under one operating system, they are essentially a cluster of systems. here you use the MPI programming model which is a bit harder to do. here you explicitly split your code to run your code in these clusters. And, the highest level of complexity in the code is that it can not only use this cluster, but use graphics processors in the cluster for computing. earlier graphics processors were only used for drawing, more and more people are looking at using them for computation also.

In the past all this has been ignored in the enterprise space because people were happy doing one processor and one memory. But, now the clock speed has stopped improving, even the enterprise world cannot ignore that now, they have to climb up that ladder.

The applications include predicting climate change in the future. We have examples of people using 10,000 cores for one such simulation for creating earth models. NASA, one of our customers, is using a cluster with 75,000 cores, to help detect planets. in science they already using thousands of cores for a single application.

A: In the cloud services space we start with the lowest level, IaaS (infrastructure as a Service) like Amazon eC2, the next level is Platform as a Service and the highest level is Software as a Service where an application is given as a service. We provide this service for some sectors like genomics, engineering Design, oil and gas and so on.

We have started this business recently and now we have a number of customers that are using our systems. Typically we build a system based on requirement. We do not have a big outlay of hardware behind that is waiting for customers. Since we are building the hardware, anytime we get an order, we build a system for them. Currently our business is at the SaaS level.

A: This is something that we have thought of for a number of years now. in order to be a highly successful company we not only have to diversify our feature set, but our market too. rackable's acquisition of SGI gave us that avenue on the cloud side. The big question is, can we use HPC-class systems in the enterprise?

one of the early successes of HPC in the enterprise world is in Data Analytics. We realize this more and more that one thing that is common between the enterprise and HPC is data handling. The common problem is dealing with massive amounts of data and making sense of it. We have started selling our big memory systems for doing fraud detection in the enterprise world. for instance, Ebay uses our systems for fraud detection in their PayPal system.  We have other enterprise customers that are doing the same but prefer not to be named. our systems are also being using for efficient deep packet analysis of  internet traffic.  

So, as we started niche in the media space before diversifying, similarly we will be starting niche in the enterprise segment also. The niche in this case will be the ability to handle large amounts of data with deep analysis of that data.

A: A certain big telco in the U.S. came to us because their main database server grew very large due to mergers and acquisitions. Whenever somebody tried to call from their phones, it was taking longer and longer for the connection to be made. A large chunk of the connection time is related to scanning the database for that connection. They are using our servers for this now, to handle that large database and do that deep inspection of data when a call is being made on their networks.

A: We are actually of great interest to the financial services industry, not only to BFSI but to high frequency traders too. These people buy fast computers to do massive amounts of trade in a very very short time. They earn their profit from very small variation in the stock price. They do thousands of transaction in a day, that cannot be done manually. They want systems to take decisions very fast at the flutter of variation. We have been very popular with those customers.