I was surprised to realise that it’s nearly 7 months since I bemoaned the lack of availability of 10gR2 on Solaris 10 (x86). Well, it’s been out a few weeks now and I’ve been waiting since then for a free weekend to install it on the ISP4400. One of the advantages of being busy the past few weekends was that it gave Howard Rogers the chance to produce a couple of excellent articles. The first is a guide to the Solaris 10 installation, the second covering the Oracle install on top.
Having worked my way through the articles, I can vouch for their usefullness and accuracy. I probably wouldn’t stick the Oracle software on the root filesystem, which I’ve mentioned to Howard, but everything else is spot-on. Thoroughly recommended.
With the time it saved me through not messing around, I was able to get straight onto creating a ZFS pool without delay. How easy is this?
# zpool create oradata c0t1d0 c0t2d0 c0t3d0 c0t4d0
# zpool list
NAME SIZE USED AVAIL CAP HEALTH ALTROOT
oradata 136G 1.10G 135G 0% ONLINE -
# df -k oradata
Filesystem kbytes used avail capacity Mounted on
oradata 140378112 24 140378033 1% /oradata
# mkdir /oradata/TEST1020
# chown oracle:dba /oradata/TEST1020 Of course, it might perform like a complete dog, but that’s what I’m hoping to find out one way or the other 😉
Updated later … Here’s an interesting link that seems to confirm what I expected, that ZFS is not the best performing solution at the moment. (Well, certainly not for OLTP workloads) It confirms the earliest tests i’ve done too.
Looks like a damn interesting blog all round, actually – added to my Blog-roll.
Perhaps you could try some of the Sponge’s test scripts for Linux disk scheduling – afterall you are now running stripe across all volumes.
I wonder how much overhead the 64-bit checksums on the data impose on IO – it’s getting a bit like raid 10 meets 5.
Still, it’s nice to have a file system that even a DWer could not fill. 😉
I wonder how much overhead the 64-bit checksums on the data impose on IO – it’s getting a bit like raid 10 meets 5.
I have to say I’m not particularly optimistic about ZFS performance, but I can see that the flexibility amd reliability might be just the ticket.
It’s going to take quite a bit of playing around to see how it goes but I must admit it’s absurdly simple (and fast!) to set up.
Excuse a thicky for asking, but what file system was I using after I’d finished installing Solaris? I forgot to check!
UFS, I’d guess
$ grep /dev/rdsk/c0t0d0s0 /etc/vfstab
/dev/dsk/c0t0d0s0 /dev/rdsk/c0t0d0s0 / ufs 1 no –
Righty-ho. Live and learn:
http://en.wikipedia.org/wiki/Unix_File_System
The following blog on ZFS is quite good as well …
http://blogs.sun.com/realneel/entry/zfs_and_databases
Andy
Nice one, cheers.
Not sure I agree with some of the content, though …
Depending on the workload and configuration, you can have thousands of IO operations per second. The size of these IO is usually small (database block size). All the writes are synchronized writes.
Looks extremely interesting nonetheless and I definitely need more info because my intial test results were atrocious 🙁