Storagebod So VSAN is finally here in a released form. On paper, it looks very impressive.
I spend an awful lot of time looking at scale-out storage systems – looking at ways to do them faster, cheaper and better. And although I welcome VMware and VSAN to the party, I don’t think I’m really the target market: the product’s not really ready or appropriate for media and entertainment or anyone interested in hyperScale.
But even so, I’ve got thoughts that I’d like to share.
Is VSAN better because it runs in the VMware kernel? This seems logical but this has tied VSAN to VMware in a way that some of the competing products have not. For example, if I want to run a Gluster Cluster – which encompasses not just VMware but also XEN, bare metal and anything else – I could. And there might be some excellent reasons why I would want to do so. Perhaps I’d transcode on bare-metal machines but might present out on VM-ed application servers, for example. Of course, it is not only the media and entertainment industry which has such requirements – there are plenty of other places where heavy lifting would be better done on the bare metal.
Open house party?
I think that VMware needs to be much more open about allowing third-party access to the kernel interfaces. It should allow more pluggable options so I could run GPFS, ScaleIO, Gluster, Stornext within the VMWare kernel.
VSAN limits itself by tying itself so closely to the VMware stack; its scalability is limited by the current cluster size. There are plenty of good architectural reasons for doing so, but some of these are enforced by a VMware-only mindset.
And why limit yourself to only 35 disks per server? An HP ProLiant SL4540 takes 60 disks and there are SuperMicro chassis that take 72 disks. Increasing the spindle count not only increases the maximum capacity but the RAW IOps of the solution. Of course, there might be some saturation issues with regards to the inter-server communication.
Yet, I do think it is interesting how the converged IT stacks are progressing. VMware itself is pretty much a converged stack now but it is a software converged stack; VCE and Nutanix, on the other hand, converge onto hardware as well. And yes, VMware is currently the core of all of this.
I actually prefer the VMware-only approach in many ways as I think I could scale computer and storage separately within some boundaries. Additionally, I’m not sure what the impact of having unbalanced clusters will be on VSAN. Would make sense to have some Big Flipping Dense VSAN appliances rather than distributing the storage equally across the nodes?
VSAN is certainly welcome in the market: it certainly validates the approaches being taken by a number of other companies – I just wish it were more flexible and open. ®