VMware's Virtual SAN Threatens Traditional Data Storage Models
VMware has launched the final part of its software defined data center puzzle: a virtual SAN product called Virtual SAN.
The product has been in beta testing for the last six months, with around 12,000 customers, but there were still plenty of surprise announcements made at the launch event on March 6.
The biggest of these was the maximum size of its Virtual SAN. Previously VMware had said that this would be 8 server nodes, and then 16. But VMware CEO Pat Gelsinger announced that the product has been upgraded to support 32 nodes.
Now for some math. Each node can host up to 32 disks and up to 100 virtual machines (VMs). That means that a 32 node Virtual SAN can host 3,200 VMs . It can also provide a maximum of 4.4 petabytes of storage space, and offer 2M IOPS (100% read) .
"This is a monster," said Gelsinger, echoing the monster VM concept that the company introduced in 2011 with the introduction of vSphere 5. "This will take care of the storage requirements of pretty much any enterprise, and they don't need to deal with any hardware underneath."
He added that performance scales linearly - a 16 node setup offers 1M IOPS, a 32 node one offers 2M IOPS - and despite his previous comment about 32 nodes being monstrous, he hinted that even larger Virtual SANs are on the roadmap: "There will be more in the future. Just you wait."
So how does Virtual SAN work and what makes it different from other virtual storage products from the likes of Nutanix or Nexenta? The key difference is that most virtualized storage systems require a storage hypervisor or virtual storage appliance running on top of VMware's hypervisor, alongside virtual machines, or some other software.
This can cause problems with resilience and performance, which VMware's solution avoids, according to Ben Fathi, VMware's CTO.
"We have built Virtual SAN into the hypervisor, so there is no special software to install and maintain. You just install vSphere," he explained.
Being embedded in the kernel of the hypervisor means Virtual SAN is uniquely positioned in the software stack for visibility into applications. It also has a unique view of the infrastructure beneath it that allows it to optimize the I/O data path to deliver better performance than a virtual appliance or external device, he claimed.
"It takes disks and SSDs and abstracts them, and pools them as a shared storage tier, and you can put VMDKs on them," he continued. "You no longer have to deal with the underlying hardware, LUNs, fibre channel - you don't need to bother with any of that."
Fathi specifically mentions SSDs, and in fact it is a requirement of every node that it contains at least one SSD. The reason for that is that the inclusion of SSD storage allows applications to be optimized to run faster, and it also allows nodes to be rebuilt faster in the case of a hardware failure. The SSDs themselves don't affect the storage capacity of each node - 30% of their capacity is used as a non-persistent write buffer, and 70% is used a read cache.
The next question is: how do you build a Virtual SAN?
It turns out that there's two ways. The first way is to buy VSAN Ready Nodes or VSAN Ready Blocks from a VMWare partners such as Cisco, Dell, Fujitsu, IBM and Supermicro. A Ready Node is a single tested pre-configured server and a Ready-Block is a tested pre-configured set of servers for use with Virtual SAN. They include controllers, disks, and SSDs and are ready to plug in and run.
The other way is to build your own nodes by choosing individual components from the Virtual SAN Compatibility Guide: any server on the vSphere Hardware Compatibility List, at least one multi-level cell SSD (or better) or PCIe SSD, at least one SAS/NL-SAS HDD or select SATA HDD, and a 6Gb enterprise grade HBA/RAID controller. (Nodes which don't contribute any storage resources can also be added to a cluster to improve compute performance for the cluster as a whole.)
How much will they cost? At the moment that’s the $64,000 question because Gelsinger didn't provide any pricing details at all. Reports of license costs leaked on the Internet suggest that a Virtual SAN license may costs around $750 per CPU, and if you estimate a server cost of about $5000 - $6000, with an extra $5000 or so for SSD and HDD drives, you get an all in per-node price of about $11,000 to $12,000.
That would make a 3 node cluster come in at around $35,000, and a large 32-node cluster would cost around $370,000.
Once you have your Virtual SAN up and running, it is integrated with the rest of the VMware software stack. That means that you can do most of what you can do with traditional storage. That includes vMotion, snapshots, clones, Distributed Resource Scheduler (SRS) , vSphere High Availability (vSphere HA), vCenter Site Recovery Manager (vCenter SRM) and more.
There is one caveat thought - there is no talk about deduplication or compression as features of Virtual SAN (yet.) These are features that other virtual SAN providers offer, and ones that many customers are likely to want. Getting those features will costs extra until they are provided in a possible future upgrade.
One of the features that Gelsinger was very keen to mention was simplicity, and the company showed that once the hardware is in place with vSphere running on top of it, it is remarkably easy to get a Virtual SAN up and running.
The first step, from vCenter, is to turn on the Virtual SAN feature. Then you need to allocate disks to the Virtual SAN pool, and this can either be done manually by selecting them, or automatically - the software finds all unused disks and adds them to the pool and adds any new ones when they come online.