On 14th of May 2012, Nutanix has officially started its EMEA division. I was testing at that time Fusion-IO cards, and I was really interested in Nutanix as one of the best implementation of the Fusion-IO “no SAN” idea. Luckily I was their official first guest, since we had a webex on that same day early in the morning. Alan and Rob showed me their technology, answered to all my questions and gave me a lot of material. This article aimes to be a recap of what I learned that day, and the first one of a series in the upcoming months.
The Company
Nutanix is a start-up company founded in in September 2009, by three cofounders. CEO Dheeraj Pandey, CTO Mohit Aron and Ajeet Singh – were all previously at analytics database provider Aster Data Systems. Before that, Pandey and Singh were both at Oracle, where Pandey was involved in the development of the Oracle Database Machine and Exadata, while Singh helped plan Oracle’s initial cloud computing strategy. Aron was at Google, where he led the design and development of the Google File System (GFS).
In the past, Google built a custom distributed scale-out architecture of converged compute and storage nodes using GFS to manage the data and ensure fault tolerance. In doing so, it saved money on hardware and boosted performance by relying more on local data access. Nutanix is leveraging a similar approach to large-scale distributed systems, this time tuned for virtual- ized workloads. This paradigm of bringing compute and storage together was the primary inspiration behind the company, along with the demonstrated success of vendors who have successfully delivered rich software services through an appliance platform.
The solution
Nutanix Complete Cluster is for me a revolutionary solution for datacenters and virtualized environments. Nowadays there is a hype on scale-out, and both servers and new storage solutions are going this way. But even when using fast solid state disks, data always have to cross the network between servers and storage, and can become a bottleneck.
Nutanix has taken a completely different approach, based on the “shared nothing” or “No SAN” concept. Basically, it’s all about taking away SAN storage (and thus the network connecting it to servers) and bring it directly inside every server: every node of the cluster has an amount of local storage, and thanks to the NDFS (Nutanix Distributed File System, the Google background here shows its value…) its data are replicated across other nodes to guarantee both redundancy and the possibility to execute every virtual machine from local storage, even after a vMotion operation.
The scale-out approach works for both compute and storage, since every nodes has both. When you add new nodes, NDFS simply replicates its data also to the new servers, so at the end a single filespace grows as needed, or as they say “grow as you go”.
From a hardware standpoint, Nutanix uses half-widht 1U servers, and a 2U chassis that can hold up to 4 nodes. Every node has its own power unit, connectors and disk, so there is no shared backplane like in blade servers, and this is to me a great advantage over competition.
Every server has two Intel CPUs, from 48 to 192 Gb of Ram, and a tiered storage starting from a Fusion-IO card down to a SSD disk a 5 sata disks. tiering and lun creation is managed by a proprietary VSA (Virtual storage appliance) installed in every node, and replication is made thanks to 10G connections in the backend.
Conclusions
Nutanix has taken a radical approach in its design, and is really different from any other competitor. The idea of using building blocks in a datacenter expansion like a LEGO system is really fascinating, and has also other advantages like rack space and power/cooling consumption. One of the use case we thought about during our meeting was colocation, where you have to pay for the rack units you use. A Nutanix system can save you big money in this situation.
A final note about pricing, since I think everybody would ask me about it. First of all, remember you are not buying blade servers! A Nutanix solution has to be compared with both servers and a tiered storage system with SSD and flash cache. With this in mind, a four node block with 48GB RAM per node and 20TB total storage is about 138.000 USD.
The minimum Nutanix configuration is three nodes each with either 48GB, 96GB or 192GB of RAM that provides a total of 15TB of raw storage (comprising of Fusion IO, SATA SSD and SATA HDD) and will push 15,000+ IOPs