On Jul 30, 2009, at 1:49 AM, Chris Hunt <[email protected]> wrote:
Any word?... How did things come out?
Which one or probably I should say whose?
I ran a Solaris 10u7 iscsitgt benchmark with horrible results against
what I think is a fantastic zpool config (7 mirrors of 15K SAS disks
and 512MB of NVRAM).
I think the major problem for me is that ZVOLs on Solaris 10u7 are
still completely async, which actually is a performance penalty for me
with NVRAM cache. Not to mention grossly negligent on Sun's part, as
an iSCSI volume on a ZVOL in Solaris 10u7 can loose up to 1/8th the
amount of memory of data in a system crash under heavy write load. Sun
should have issued the change of that as a critical fix ASAP instead
of waiting or Solaris 10u8 (it wa initially promised for Solaris 10u7!).
No, that was the straw that broke the camel's back for me. No matter
how good the file system is, if the supporting services aren't up to
the task, then the system isn't up to the task. The lack of file
system repair tools for ZFS was also a major factor. The arrogance of
the developers to think their file system was bullet-proof. There is
no such thing as bullet proof.
After the lack luster experience I'm going to have to retreat back to
Linux and hardware RAID. I may look again later when things improve:
file system repair tools, ZVOLs that behave properly both for writes
and reads, better iSCSI target software, but in the meantime I'll
stick with the reliable and predictable performance of HW RAID even if
there is the remote possibility of silent disk corruption.
-Ross
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss
