Raw vs. cooked

Let's put this issue to rest -- or maybe not: Is IDS's performance better using raw diskspace or the Unix filesystems, aka "cooked files?" A Sun field sales/system engineer recently told me that Sun's filesystems are so sophisticated now that the performance is identical. Another system developer told me that he had benchmarked the two, and had seen a slight disadvantage to using Informix on Solaris with raw space. However, he said a different Unix flavor had abysmal cooked file performance compared to raw. What have others found? Does Informix have any published (recent) benchmarks that address this question?

Tom Smith wrote: > > Let's put this issue to rest -- or maybe not: > > Is IDS's performance better using raw diskspace or the Unix filesystems, > aka "cooked files?" A Sun field sales/system engineer recently told me > that Sun's filesystems are so sophisticated now that the performance is > identical. Another system developer told me that he had benchmarked the > two, and had seen a slight disadvantage to using Informix on Solaris > with raw space. However, he said a different Unix flavor had abysmal > cooked file performance compared to raw. > > What have others found? Does Informix have any published (recent) > benchmarks that address this question? I have not done any formal testing in a few years but it always comes up the same: RAW is 15-25% faster than COOKED on every OS I have ever tried no matter how "sophisticated" the FS is. Let Sun show us the benchmarks, fully documented. Art S. Kagel

Using AIX 4.1 and DynamicServer 7.2 on a four processor RS6000, I found cooked beat raw by a margin of 15%. This was running against Clarion array disks, the raw was non-raid, the cooked was raid-5. The local Informix people checked my results and went away shaking their heads.