My issue is not directly related to physical hard disk performance (even if that is another point), but on RAMDISK performance.
Devsk is right: I am modelling memory performance. A ramdisk without a file system on it (such as the one obtained using the ramdisk create command) is like a raw disk. Since no zfs is on that, I do not think that deduplication or compression apply to this case. Now, the figures point out: 1)Ramdisk performance is around 658MB/s. Devdsk is right: ddr3 performance should be around GB/s not some hundred of MB 2) To facilitate comparison I posted an even worst case, which is a 16 physical disk array with a file system (ZFS) on it: 578MB/s. Now it seems to me quite strange that a raw ramdisk has a similar performance to a pool of physical devices with zfs on them. My second point is that even the performance of the disks is poor: theoretical stripe performance is 3200MB/s. Ok: we have controller (a 6Gb/s SAS2), RAIDZ3 and ZFS, but I do not this this justifies such a performance drop.... So something is wrong somewhere .... at possibly is low level and shared between ramdisk and mpt2 SAS ... Do you agree ? Any idea ? -- This message posted from opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
