<   Previous Post  Storage Virtualizati...
Half term - R&R  Next Post:   >

Comments (2)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 localhost commented Trackback

You are learning well, grasshopper. Maybe too well, even.<div>&nbsp;</div> Sadly, you're still stuck in the trap of comparing configurations without any real regard to cost, application diversity, replication overhead or the reality of drive rebuilds.<div>&nbsp;</div> Though you point out that cache hostile workloads can theoretically be overcome with sufficiently large cache, that too is a balance of spindles, dataset size and the effective locality of reference/access density.<div>&nbsp;</div> And somehow I can't buy into you're belittling the importance of 1.1-1.2 milliseconds of response time, especially when you boast of a mere 60 microseconds of overhead for a read miss. Although the SVC's overhead may be a fraction of the total time, 1 ms better response time is massively significant when we're talking 2ms vs 3ms...<div>&nbsp;</div> Seems you may be adapting to this imaginary math too well, my friend :) <div>&nbsp;</div> But to be clear - my point remains that the SPC-1 is a totally meaningless tool to compare different storage platforms with different configurations, or to compare different configurations of the same storage platform. There are just too many unknown variables that will impact observed performance...to meet the scientific method, there are numerous additional variables that must be removed, and implications of many real-world influences that must be introduced back in before this is in any way meaningful.<div>&nbsp;</div> Until then, you're all just poking holes in each others' paper pinatas...

2 localhost commented Permalink

BarryB,<div>&nbsp;</div> While I agree that 1ms can make a big difference, the point I was trying to make was that with SVC using mid-range storage in this test, its not really surprising that there is a slightly higher response time. If we were to stick several fully populated DS8000 behind SVC and run the benchmark again I'd expect the numbers to be comparable if not better. The $/IOP would likely be much more.<div>&nbsp;</div> I have had to amend my post, based on the general retraction of the comments that 'theirs is faster than ours' but however, the rest of the statistics are still posted and so I have left my rationalization of these in place.<div>&nbsp;</div> Thanks for you thoughts as always.