Slower MySQL inserts for AMD64/Opteron?

ray at redshift.com ray at redshift.com
Thu Jun 23 18:50:24 GMT 2005


Thanks, I will do that :-)  When you speak about IDE/ATA, does that apply to
SATA as well?  Or just PATA?

Ray


At 06:43 AM 6/23/2005 -0700, David O'Brien wrote:
| On Wed, Jun 22, 2005 at 10:41:59AM -0700, ray at redshift.com wrote:
| > I did some benchmarks recently on a single 2.4Ghz Xeon against a Dual
| > 246 Opteron and noticed when it came to MySQL inserts, the single Xeon
| > was about 20% faster on inserts (and it actually had a slower hard
| > drive, IDE vs SATA).  The interesting thing is that the Dual Opterons
| > was twice as fast retrieving the data using selects.
| ..
| > The Xeon was running 5.3/i386.  I believe the AMD was running 5.4/AMD64.
| 
| Note that in 5.4, the ATA subsystem runs without the Big-GIANT-Lock.  CAM
| (SCSI subsystem) still grabs GIANT.  For a system with a single activity
| going on (such as yours), I feel ATA can "out perform" SCSI doing some
| things.  Note that most IDE disks lie and always cache things.  If MySQL
| is doing fsync's to ensure the data is on disk, the ATA disks can seem
| faster, but aren't as save and reliable than SCSI disks.  There are many
| reasons one uses SCSI over IDE disks.  "Raw" performance isn't
| necessarily one of them.
| 
| Summary: please try an IDE disk in the Opteron machine.  Preferable pull
| the disk out of the Xeon, put it in the Opteron and re-run your
| benchmark.
| 
| -- 
| -- David  (obrien at FreeBSD.org)
| 
| 


More information about the freebsd-amd64 mailing list