Q&A with Tiffany To of Nutanix - Page 2

By Tiffany To (Profile)
Monday, February 27th 2012

VSM: One of your co-founders came from Google, and I’ve heard Nutanix say the product is "Google-like". What does that mean?

TT: Cloud-Generation companies like Google, Yahoo, Microsoft (Azure), and Amazon realized traditional datacenter architectures would not meet the needs for massive data sets, high performance at a profitable cost model. Google pioneered the first cloud-generation architecture with a distributed scale-out approach of bringing compute and storage together with commodoity hardware to eliminate the network bottleneck. Nutanix’s CTO and Co-Founder, Mohit Aron, led the design of the Google File System, which enabled Google’s application developers to harness the power of their cluster, and ensure fault tolerance for the commodity hardware that was used. This converged infrastructure innovation not only saves Google on cost, but delivers higher performance by enabling largely local data access, and became the inspiration for the Nutanix architecture. Nutanix leveraged many of these GFS architectural elements in its Nutanix Distributed File System, which is at the heart of Nutanix's clustering technology. Unlike traditional distributed filesystems, NDFS intelligently localizes blocks of data on same host where they are being accessed, delivering motherboard bus speeds without sacrificing data integrity.  It combines striping, replication, auto-tiering, error detection, failover and automatic recovery to create a true scale-out system for virtualization.

VSM: Many customers are trying to figure out the right way to use flash in their datacenters. You mentioned Nutanix uses Fusion-io ioMemory in your box. How does this deliver performance for virtualization?

TT: Today, customers can choose from having flash connected to their servers or packaged into storage arrays. The problem with the latter approach is that you end up paying a lot of money for performance that is then throttled by layers of networks switches and array controllers. A single Fusion-io card would be able to easily saturate all the bandwidth of a 10GbE or FC network, so customers would be paying for performance they can’t actually use. If they put the card on a server in an existing architecture, though, customers achieve higher performance, but lose out on key features like high availability, snapshots, data tiering, disaster recovery, etc.

The Nutanix philosophy on flash is that you need the best of both worlds, high performance with enterprise features. Each Nutanix Complete Block contains four Fusion-io ioDrives that are PCI-e attached, bringing this performance to the point of compute where the virtual machines are hosted. The Fusion-io ioMemory is also used as a Heat-Optimized Tiering cache (HOTcache) so that frequently accessed data resides here but is then automatically migrated to SATA drives when the data becomes cold so that performance is delivered on demand when needed. This approach enables customers to reap the benefits of flash at a cost closer to hard drives, and not give up anything on the enterprise feature set.