Hyperconverged Systems Go Mainstream
In a way, Hewlett-Packard, through its LeftHand Networks storage acquisition, was one of the pioneers of the hyperconverged systems market, although it didn't talk about it at the time that way. And the company is now going to mainstream the idea in a big way, using its own virtual SAN software as well as that from VMware, along with a number of other players who are also jumping on VMware's EVO:RAIL bandwagon as a means of combatting the hyperconverged upstarts such as Nutanix, SimpliVity, Scale Computing, Pivot3, Gridstore, Maxta, and others.
With hyperconverged systems, a server cluster is designed to run virtualized server workloads and a virtualized storage area network at the same time and one the same iron rather than have them be on two distinct sets of iron and linked by a network. Nutanix, which gets billing as one of the pioneers of the idea of creating a virtual SAN cluster and running compute jobs on top of it, was founded in 2009, a year after HP acquired storage array maker LeftHand. That same year HP launched virtual SAN implementation of the LeftHand software and put it on its BladeSystem blade servers so customers could create their own hyperconverged systems, although it did not call it that. And even when Nutanix came out of stealth in the summer of 2011 with its first server-storage hybrid appliances, it was more focused on "banning the SAN" than it was touting the fact that the compute and storage could be resident on the same iron. Now, for certain kinds of workloads, hyperconverged is the way to go. The box counters at IDC reckon that the market for integrated systems – which generally do not include integrated and virtualized SAN software – will grow from $5.4 billion in 2013 to $14.3 billion in 2017, and the explosive growth of hyperconverged systems will help drive the growth of this broader market.
The fact that VMware has launched its own hyperconverged platform, called EVO:RAIL and based on its VSAN software running on very strict systems configurations that server partners must hew to, will no doubt help mainstream hyperconverged systems. The fact that HP is supporting the EVO:RAIL effort as well as offering up a LeftHand alternative that allows customers more freedom of configuration than VMware is allowing should go a long way towards making hyperconverged systems more normal. HP is still the volume leader in servers and is also the volume leader in integrated storage arrays on servers, so it stands to reason that HP's renewed efforts in this arena will help it grow. For its part, Dell has partnered with Nutanix and is getting ready to ship its own clusters using the Nutanix Operating System and the Nutanix Distributed File System and is also peddling EVO:RAIL systems. VMware parent EMC, Supermicro, Fujitsu, Hitachi Data Systems, Inspur, and Net One all have committed to ship EVO:RAIL clusters, and it can't be long before Lenovo (which just bought IBM's System x server division) will join the ranks.
HP has been distributing a license to the LeftHand VSA software with every ProLiant Gen9 server; this license is capped at a 1 TB virtual SAN, but it gives customers a feel for what the software can do. To date, HP has shipped over 1 million licenses to the VSA software, says Rob Strechay, director of product marketing for software-defined storage at the HP Storage division. HP has tens of thousands of customers who are running VMware ESXi hypervisors on top of the VSA storage, and while there are a great number of physical appliances there, Strechay says that the majority of those ESXi stacks that are running atop VSA are running on the virtualized storage. In this sense, one can argue that HP has already mainstreamed hyperconverged systems, although it did not do it in any formal way.
This week, HP got formal with the launch of two sets of hyperconverged appliances, one running its VSA and the other adhering to the EVO:RAIL specs from VMware.
The hyperconverged systems using HP's homegrown VSA software are called the ConvergedSystesm 240-HC StoreVirtual and 242-HC StoreVirtual. They have slightly different configurations based on performance needs for the storage, with the former being all disk and the latter mixing in flash-based SSDs. Both machines are based on HP's existing SL2500 hyperscale chassis, explains Jeff Carlat, director of product management for converged systems at HP, and use Intel's prior generation of "Ivy Bridge" Xeon E5-2600 v2 processors in the ProLiant SL-200 Gen8 series of server nodes. The SL2500 chassis has four server nodes in an enclosure, and HP is able to cluster up to eight of these enclosures, for a total of 32 nodes, into a single StoreVirtual VSA.
On the CS 240-HC StoreVirtual machine, the nodes have two Ivy Bridge Xeon E5s with eight cores each running at 2 GHz and 512 GB of main memory. The chassis has a total of 24 1.2 TB SAS drives in a 2.5-inch form factor and has either Ethernet ports running at 10 Gb/sec and another eight ports running at the much slower 1 Gb/sec. When the overhead of RAID 10 and the VSA software is taken out, each chassis delivers 12 TB of usable capacity, for a total of 96 TB over the 32 nodes in the fully-extended cluster. Customers who need a little more oomph for their workloads can opt for the CS 242-HC StoreVirtual machine, which ramps up to ten-core Xeon E5 v2 processors running at 2.8 GHz, 1 TB of main memory per node, and which swaps out eight disk drives for 400 GB SSDs. The nodes in the chassis are linked to each other through a backplane, and the 10 Gb/sec Ethernet ports are used to link the nodes to each other and to the outside world of end users through a top-of-rack switch. All of the nodes are certified to run VMware's ESXi server virtualization and vSphere management extensions, and Carlat says that HP is working on a variant of the StoreVirtual appliance that will support Microsoft's Hyper-V and the open source KVM hypervisors in addition to ESXi. HP has integrated its OneView Instant On management extensions with ESXi and StoreVirtual so is can automatically deploy both sets of software on new nodes as they are added to the cluster.
These two hyperconverged systems using HP's StoreVirtual VSA as storage will ship in December; pricing has not yet been set.
In the first quarter of 2015, HP will debut an EVO:RAIL setup based on the ConvergedSystem 200-HC platform, The exact feeds and speeds of that cluster were not divulged, but they have to meet VMware's precise EVO:RAIL configuration requirements, which leave little room for differentiation. And intentionally so. VMware wants to minimize its support headaches with both ESXi and VSAN, and that means locking down the configurations. While VMware's VSAN scales to 32 nodes, like HP's StoreVirtual VSA, VMware is limiting EVO:RAIL to 16 nodes at this time. (You can see some initial EVO:RAIL setups that EnterpriseTech saw at VMworld in late August at this link.) VMware has its own management stack for the EVO:RAIL clusters as well and OneView tools will not be necessary.
The question now is what one will HP customers go for? The homegrown HP setup or the EVO:RAIL?