I always love doing John Troyer’s podcasts – the one we did yesterday was the one discussing the iSCSI multivendor post we did (here). You can listen to a recording of the podcast – here.
A big thank you to Eric, Adam, Vaughn and Andy for the work on the post and on the call.
On another note….
I’m always a bit frustrated when the “what protocol is best” or “which protocol is tier 1” question comes up. Us poor human beings, we like to see the world as black and white, but it just doesn’t work like that.
Answer we gave on the call (and hearing Adam from HP/Lefthand, Eric from Dell/EqualLogic, Vaughn from NetApp and I from EMC all say in essence the same thing) should tell you something.
- iSCSI, FC, NFS – they all work, and done right, they can all support a broad performance envelope, and a broad availability envelope.
- The real answer varies by customer, and requires you dive into the details.
- There is varying maturity – both inside the customer and in the industry and technology stacks. For example, with a strong MPIO implementation (like in vSphere), iSCSI’s ability to support a set of workloads immediately becomes much, much broader than it was in VI3.x But more important than all of that is what do you already have, what do you have experience/expertise in.
- Each protocol has different design considerations (as a focus area). Example: for block devices, you need to think about queuing and MPIO; for NAS devices it’s “single TCP session” and failover timeout values
- Each protocol has unique positive values Example: iSCSI in guest or pRDMs on block can be used for specific app use cases (clustering, V-to-P emergency “repro in a physical” scenarioes); file-level or VM-level snapshot/mobility operations are architecturally easier to implement on a NAS device than they are on a block device.
- If you’re deploying IP storage, you need to approach it with the same rigor and end-to-end thinking/approach as any traditional FC SAN – particularly if you’re not just playing in a lab, but running a production environment. This means (yes) that if you’re a shop where you don’t have end-to-end control of VMware, networking and storage – you need to work as a team in your enterprise.
Please stop asking the question? Not because I’m not happy to answer, it’s just that there are so many better questions, IMO, and the choices you have a a customer mean that you don’t need to look at this as black and white.
I’ve always been consistent – the most flexible scenarios use block and NAS, and the choice of what block (FC, FCoE, iSCSI) is up to the customer’s existing infrastructure/expertise. Converged fabrics make this question a non-question (as you can have your cake and eat it too).
Comments