Since last year, HP rebranded its LeftHandstorage with the new name StoreVirtual, while the hystoric name LeftHand has been left to identify its operating system. While the relase of the new 11.0 version of the software is imminent, HP has just released the new official document titled What’s new in LeftHand OS 11.0, worth a read to learn about the new upcoming features, and if you already have a StoreVirtual VSA, be able to plan the update pondering the use of the new features it brings.
There is no doubt the most awaited feature is Tiering. HP calls it Adaptive Optimization, and allows the transparent use inside the StoreVirtual VSA of two different layers of storage with different performance levels (mainly SSD and HDD), by automatically moving sub-blocks of the same LUN between the two tiers, in real time and with a granularity of 256 KB, that is the usual LeftHand page size.
While is other traditional storage systems with tiering, where the data moves between tiers on a periodic schedule based on hystorical performance data, in StoreVirtual tiering is real time. I’ve never been a fan of “old style” tiering, because I’ve always felt it was not able to manage I/O spikes, instead it was always late in adapting to I/O requests. Take this example: there is a virtual machine with a software responsible to create wage payments for a big company: this VM is almost idle for the whole month, but it has a sudden spike of I/O the day the payments are made. In a storae with scheduled tiering, we can expect the blocks of storage where this VM is saved are on the cold tier. In order to move it to a higher tier, the storage first need to realize the I/O is increased: but since the tiering algorythm is based on hystorical data (usually the previous 24 hours) it will see the I/O increase at least 24 hours later, when probably the monthly activity would have already finish. You can imagine the final result: the virtual machine is moved in the performance tier too late, making the tiering itself completely useless.
The only way to prevent those problems in traditional systems is to schedule the tiering moves in advance, but obviously this solution is prone to several problems: even a simple date change in the I/O spike makes again the schedule completely useless. Or, you can stick the VM in the higher tier, but this in turn has an impact on budget because of the wasted disk space on an expensive tier.
so, real time tiering like this one in HP StoreVirtual is more than welcomed. StoreVirtual has always been a good and solid product, but it was starting to show its age, especially on the performance side. They only way to increase performances was to create bigger and bigger stripes and clusters with more nodes. And it was not the ideal solution, since you were trying to solve a performance problem with a capacity solution. Tiering is a great new feature, and it moves again StoreVirtual in par with other “modern” solutions.
HP StoreVirtual VSA will be available in three different versions:
The “Basic” version is really appealing for all those customers that bought or are planning to buy the vSphere Essentials (Plus) bundle from VMware: it allows the deploy of a VSA per ESXi host. there is a capacity limit at 4TB per VSA, this means a Network Raid 5 lun can be as big as 8 TB, and is a size usually enough for SMB customers. Also, it does not offer auto-tiering.
Another interesting version is the da 50 TB, called “Capacity”: VSA now is the biggest StoreVirtual edition. An interesting note about the number of nodes per cluster: until today the official limit was 32 nodes, but HP itself was suggesting a limit of 16 nodes, otherwise replication traffic would have been to much, and with an impact of performances. With the new version, the “real” limit of 16 nodes was simply “certified”.
For Hyper-V users, now HP StoreVirtual supports the SMI-S protocol; this means you can manage volumes activities directly inside Microsoft System Center Virtual Machine Manager (SCVMM) 2012 SP1, without the need to use HP Centralized Management Console.
Also, there are new and improved versions of several components: CLIQ (the command line interface), snapshots support when using Microsoft VSS, and HP StoreVirtual Recovery Manager.
Finally, the way snapshots work has been slightly changed: if a scheduled snapshot does not uses additional space, its removal is postponed even if it’s expired, but it’s kept in the cluster; this saves performances without sacrifying space. You can see those kind of snapshots in CMC (and via command line) with a specific tag, named “consolidating”.
the new features coming with LeftHand 11.0 are really interesting, and obviously the biggest one is Adaptive Optimization. And besides the benefits it brings, in my opinion it also brings a statement about the future direction of the whole StoreVirtual family: tiering and new sizes will be available with the VSA, to me sounds like the future of StoreVirtual is only in its software versions, that is the VSA.