I was invited to a briefing by the vendor Nutanix on Monday at VMworld. Now there are a lot of new / recent startups in the storage space and keeping a handle on them all could occupy my time completely so I did hesitate to accept the invitation at first.
I had heard some good things about Nutanix from other bloggers though and, after looking at their website, I was intrigued to find out a little more. Along with a few other bloggers I found my way to the Tryp Apolo hotel in Barcelona where we were greeted by a number of Nutanix employees from EMEA and the US along with London VMUG’s very own Jane Rimmer.
Perhaps now is a good time to explain what it is that Nutanix do. They claim to be a software company but their software is only available on their hardware. I would perhaps think of them more as a storage solutions company. Anyway, that’s semantics.
Nutanix’s product aims to provide a full virtualization platform that performs consistently well, scales linearly and, most importantly, does not requires any shared storage. That’s right, no shared storage. No SAN.
Each node (host) is a fairly standard x64 architecture server with dual processors. Presently each node comes equipped with 320Gb of PCIe SSD (fusionio), 300Gb of SATA SSD and 5Tb of SATA HDDs. Each node also has 1x10GbE and 2x1GbE networking connections. Nodes are manufactured in blocks of 4 and each node has VMware ESXi pre-installed on it.
Aside from combining the hardware, Nutanix’s secret sauce comes in when it comes to presenting that local storage to ESXi. When the nodes are clustered, the available storage is combined and presented as a VMFS datastore to all of the hosts in the cluster. VMs provisioned on a host will have their files stored locally although it will appear like they are being stored on a shared datastore when viewed through the vSphere Client. Behind the scenes the Nutanix software actually replicates those files to other hosts within the cluster (imagine that there are more hosts than shown below – this was just a quick diagram that I knocked up):
The fact that the datastore is presented to all hosts means that vMotion and HA both work as intended. If a VM ends up on another host Nutanix will move that VM’s files to the correct host in the background and completely transparently.
With respect to scaling, Nutanix say that you can just add blocks to an existing deployment. As each node has its own storage, each node should have more than adequate storage performance to handle the VM load placed on it. Clever stuff but does it really work and does it really scale?
Being the diligent bloggers that we are, we asked plenty of questions and Nutanix seemed to have all of the right answers. For me, the idea of scaling in that way is perfect for a growing business. More established enterprises may be too heavily invested in existing technologies to consider it though. Technically it’s a clever solution too, no doubt about that, but perhaps they may need to introduce a few more sizing options for the hosts over time or the software up to being used on other hardware platforms.
After that, Nutanix gave us some insights into the future development of their product. I can’t go into details unfortunately but I look forward to seeing how they progress.
Thanks to Jane and Nutanix for organising the session (and the drinks afterwards) and talking with us all.
Michael Poore is a Senior Consultant for Virtual Clarity, a small virtualisation / cloud consultancy based in London and San Francisco. Michael works on all aspects of datacenter virtualisation, automation, orchestration and management for various global companies. He started the vSpecialist blog in 2008 and convinced co-author Jeremy Bowman to join in over a beer a while later.