Benchmarking storage results, part 1
This is a followup post to last Friday’s message dealing with the hardware I have evaluated.
Benchmarking is not as much fun as I hoped it would be (or as it had been in the past). Numbers don’t always lead you to the definitive winner either as there are other factors that sometimes come up. In fact, sometimes the numbers can surprise you in ways that make you go hmmm. But numbers alone aren’t the end.
Does it perform within the parameters set (see below)? Can I measure objectively? What changed parameters caused differences in expected output? Are all features included or must I make a compromise? Is it easy to manage? Can I expect someone else to manage this? Does it fit with my long-term goals? Can it scale with my needs? Is there anybody out there?
Must be able to sustain reasonable I/O latency under load. This is one area where the current Nexenta system fails miserably at. This is probably a fault of hardware over software though local access has shown me that latency is an issue outside of the local system regardless of the NFS client involved.
Must be able to self-diagnose problems. Leaving me in a lurch will get you made fun of publicly and in general just piss me off.
Should be able to report metrics of operation over standard gathering interfaces. Nexenta has a standard Solaris MIB in place but it’s broken in a few ways such that reporting of storage usage is haphazard and inaccurate at times, or a OID just disappears randomly only to return with different usage and totals.
This was actually an easy one with all vendors coming in with quotes from just shy of $60,000 to a high of $80,000 for clustered hardware and support. This is more expensive than the Nexenta solution I built (< $40,000) but I also hope that the extra money spent gives me a return on investment in interacting with engineering and support. (In fact, NetApp just reached out middle of last week wanting in on the action though I am finding their solutions to still be more than my budget allows. Unfortunate but expected. I’ll continue to use NetApp where I need NetApp and use more cost-effective solutions for our virtualization platforms.)
I know I am sounding very negative about Nexenta. I tried to save money by going with this route over saving time by going with a more expensive solution and I paid for that mistake. I have spent far more time managing and maintaining this solution than I would have using other solutions.
For many people (companies and enterprises included), an employees time is something that’s used for this kind of work. This is not the case for myself or my employees at ipHouse (or my earlier 2 businesses).
Next up? Part 2; where I share data I gathered.