Sunday, September 30, 2012

SQLIO performance test

I always love SQLIO for disk access performance test. Recently with the arrival of a Intel 520 SSD, I started to feel curious about its performance improvement.

Below is how I test it.

The command:

SQLIO -kR -t1 -s30 -f64 -b2 -i64 -BN i:\testfile.dat >> test.txt

Parameters description:

sqlio v1.5.SG
1 thread reading for 30 secs from file C:\RamDisk\Temp\testfile.dat
using 2KB IOs over 128KB stripes with 64 IOs per run
buffering set to not use file nor disk caches (as is SQL Server)
size of file C:\RamDisk\Temp\testfile.dat needs to be: 8388608 bytes

Test result.

Disk NCQ Environment IOs/sec MBs/sec Comments
Seagate 7200RPM No Host, Windows server 2008 R2 484 0.94 ST3750640NS
WD 7200 RPM Yes Host 7,504 14.65 WD1002FBYS
WD VelociRaptor 10k RPM Yes Host 13,782 26.91 WD1500ADFD
SSD Intel 520 N/A Guest, Windows server 2008 R2 7,747 15.13 Hyper-V 2.0
SSD Intel 520 N/A Guest, Windows server 2000 20,244 39.54 Hyper-V 2.0
SSD Intel 520 N/A Host 12,646 24.70 Hyper-V 2.0
RAM DISK N/A Guest, Windows Server 2000 20,622 40.27 Dataram RAMDisk v4.0
RAM DISK N/A Host 106,945 208.87 Dataram RAMDisk v4.0

  • When NCQ is enabled for traditional hard disk, SQLIO is not reliable any more;
  • RAM DISK is still much faster than SSD
  • Windows server 2000 is much faster than windows server 2008 R2    :-)
  • Putting the whole guest OS into RAM Disk doesn't help.

No comments:

Post a Comment