Nutanix, one of the leading providers of hyperconverged infrastructure, has executed a definitive agreement to acquire PernixData, one of the leading providers of local flash-based acceleration for storage devices.
This helps to explain the number of recent exits from PernixData, which is worrying from a personnel perspective, but it does point to a pure IP grab by Nutanix. From a pure business perspective, I can understand this. There is no need to duplicate management layers, sales teams, and support: just cherry pick and train up your existing people to sell the new products, or subsume the product in your already existing portfolio. From the perspective of the people, many of whom I personally know and admire, who have invested their time and energy into the purchased company, it sure sucks.
CTO Frank Denneman intimated that he left PernixData as he could see no position that excited him in Nutanix, which in itself is worrying, as Frank is at the top of his game.
Is This a Good Acquisition for Either Company?
For those of you who have been living under the proverbial virtualization rock for the last four years, PernixData is arguably the provider of the most advanced storage offload accelerator for the VMware space. I say “advanced,” as it is the only vendor that provides the ability to accelerate machine writes as well as reads.
Nutanix is one of the leading providers of HCI, with a sideline of a boutique hypervisor called AHV. Its raison d’etre is to simplify infrastructure by providing a product that contains storage and compute in a four-node unit. One of its other main claims to fame is that due to its design, it has a preference of data locality: i.e., where the virtual machine is running, there will be a full copy of the machine’s virtual disk.
Now when I first heard about the acquisition rumor, the whole concept baffled me. Where is the synergy? Where is the hole in the buyer’s armory to make it worthwhile?
If you have data locality, then why do you require a flash acceleration product? Your data is as close to the compute as it can be. Its CVM (controller virtual machine) utilizes local storage, spinning rust for storage, and SSD for acceleration (yes you can get an all-flash version too). These CVM’s communicate with each other to keep replicated data in sync. Data in motion is kept at the SSD layer for performance and locality purposes.
This is one of the things that gives Nutanix its performance. So why PernixData? I had a lightbulb moment. Nutanix also has storage-only nodes. Its NX-6035s node only provides scale-out storage to the Nutanix Distributed File System. Now, if you have a virtual machine that is located on this device, then you obviously do not have data locality to your compute. In fact, you are almost in traditional SAN-based country. Perhaps you could just use your storage-only nodes to store your replicas, but then if there is a failure of your running node, obviously there would be a performance hit with the recovered guest.
This is where PernixData could come in. Apart from the obvious issue that FVP (PernixData’s product) is vSphere only, there is some synergy here. I am no software developer, but if FVP is re-engineered to accelerate AHV-based virtual machines, then that is a significant win. On the basis of my limited research into who is leaving or being pushed out, it seems that engineering staff is being left alone. It is mainly PernixData’s sales teams (both technical sales and actual sales) and back-office staff that are hemorrhaging, leaving the core engineering team in place.
If this is the case, this deal starts to make a little sense to me. Re-engineering FVP to run on AHV is a big win for Nutanix.