MySQL and Sun, already misleading the public...

So, someone sent me this article on MySQL and ZFS today. Not sure when it was published, but has no comments on it, so I'm assuming its recent. All I can do is laugh.

Why does Sun continue to try to compare ZFS to UFS? UFS is absolutely horrid for I/O because of its caching design. Comparing ZFS to UFS is like comparing your new brand new 2008 Mustang GT to your old 1992 Geo Metro. Yeah, both have 4 wheels and get you someplace, but not much else to compare. And to make matters worse, ZFS in many cases doesn't even beat UFS in their tests. In order to beat UFS they need to use the latest ZFS in OpenSolaris (not the ZFS that is shipping in Solaris 10).

Let me put it simply. ZFS is horrid for databases. Just horrid. Maybe if Sun would get their head out of their ass and allow you to actually do direct I/O on a per file basis (as does VxFS) or even on a filesystem basis. As it stands, try running a heavy Oracle database on ZFS, ZFS starts to push Oracle out of physical memory (even with ARC caching limited, I believe it has something to do with the wacky memory usage that Oracle does). What, do they think they are Linux now? To be honest, my biggest gripe with Linux FS's has been the overly aggressive file system caches. ZFS takes that to the Solaris world (along with rumors of NetApp patent infringement, but thats a different story).

I really want to see the MySQL folks redo that article comparing ZFS to a properly configured and installed VxFS filesystem.

I'd also get into how the author makes stupid comments regarding prefetching "wasting" time on the disk, but he obviously hasn't dealt with proper performance tuning of RAID disks that use stripe technologies with block size. Or even with the block size and record size. He has made it overly simplified in order to make ZFS sound so wonderful. Just because your database writes in a 16k block doesn't mean the average number of blocks it reads and writes at once is 1.

Oh well, what do I know...