upgraded faster SSD, slower reads

Wolfgang S. Rupprecht wolfgang.rupprecht at gmail.com
Thu Nov 1 13:28:07 UTC 2012


I just bought a fast Intel series 520 SSD (240 GB, 550 MBytes/sec) to
replace my slower series 320 SSD (120 GB, 250MBytes/sec).  I was
expecting the speed when reading the ssd to jump from my current read
speeds of around 250 MBytes/sec to (hopefully) something at least twice
that speed.  The specs say the 520 does 550 MBytes/sec for read and
almost that for write.  To my surprise the series 520 was only ~130
MBytes/sec.  This is observed under Fedora-17 using "dd if=/dev/sdb2
of=/dev/null bs=64k count=1000".  The board is an older Asus M3A78T and
has been happily running the 320 at full rated speed for 1.5 years.
 
Intel series 320:
 
dd if=/dev/sda2 of=/dev/null bs=64K count=1000
1000+0 records in
1000+0 records out
65536000 bytes (66 MB) copied, 0.254144 s, 258 MB/s
 
Intel series 520:
 
dd if=/dev/sdb2 of=/dev/null bs=64K count=1000
1000+0 records in
1000+0 records out
65536000 bytes (66 MB) copied, 0.494915 s, 132 MB/s
 
These number stay relatively constant, no matter how many seconds of
transfer I do.  Ditto for changing block size (bs=...) up to 1MByte.  I
see these numbers for the 520 when I attach it on the same SATA
3Gbit/sec link as the 320 or if I attach it on a new SATA 6Gbit/sec
controller (Highpoint Rocket 620, a Marvel 88SE9125 PCIe SATA 6.0 Gb/s
controller).  I do see it attach at 3 Gbit/sec on the old controller and
6 Gbit/sec on the new controller.  (So it isn't going down to SATA-1
speeds as far as the printf's I can see indicate.)
 
Is the 520 defective?  Ideas?  The reviews of the 520 all show it
really truly doing > 500MBytes/sec.  These suckers can really fly.
Just not for me...

Google searches show that some folks running MS software have had
similar problems and they managed to fix things by downgrading their
BIOS setting to use IDE settings for the SSD.  Other hits mentioned
something about power management on the SATA being the culprit and IDE
drivers nailed the power on instead of letting it flap in the wind.  Is
this something I can turn off from userland without recompiling the
kernel?

-wolfgang
-- 
g+:  https://plus.google.com/114566345864337108516/about


More information about the users mailing list