|
|
|
|
| Author |
Message |
onetimeonlyposter
GURU
Joined: 06 Nov 2004
Posts: 832
|
 Posted: Sat Dec 23, 2006 9:18 pm Post subject: Raid benchmark question |
 |
|
I'm new to raid and to sas drives, but have been working with parallel scsi
for a few years.
I just put up a new server with 4 Seagate ST373454SS SAS drives - supposed
to be pretty much the fastest things available. They are in a raid 10 setup
on an adaptec 4805 hardware raid controller (PCIe).
I tried harddrive tach 3.x on the system, which reports initial read speeds
at the outer portion of the drives/array around 220-240 mb/sec, dropping
eventually to 140 mb/sec, with an average of about 195 mb/sec. ...burst
speeds are almost 500 mb/sec.
This can't be right, or at least isn't measuring the same thing as what
would be measured with a single drive. I'm figuring that the max speed of a
single drive to be around 65-75 mb/sec, and Raid 10 should double that
figure (not multiply by 4). The older hard drive tach software worked on
the machine, but provided results that were completely incorrect.
Is any of this meaningful, or won't this software work correctly with
hardware raid. Are there any better benchmark tools that are better?
Thanks
Jeff
--
Posted via a free Usenet account from http://www.teranews.com |
|
| Back to top |
|
 |
|
|
Mike Walsh
Guest
|
| Posted: Sat Dec 23, 2006 9:37 pm Post subject: Re: Raid benchmark question |
|
|
It is theoretically possible to read from a 4 disk RAID 10 array at close to 4 times the speed of a single drive. Only half the data needs to be read from each mirrored drive, so they can be read almost as if they were striped, with each drive reading alternate blocks of data. This depends of course on the block size and the RAID adapter firmware. This does not apply to write operations since each drive on the mirror must write all data.
Jeff wrote:
| Quote: |
I'm new to raid and to sas drives, but have been working with parallel scsi
for a few years.
I just put up a new server with 4 Seagate ST373454SS SAS drives - supposed
to be pretty much the fastest things available. They are in a raid 10 setup
on an adaptec 4805 hardware raid controller (PCIe).
I tried harddrive tach 3.x on the system, which reports initial read speeds
at the outer portion of the drives/array around 220-240 mb/sec, dropping
eventually to 140 mb/sec, with an average of about 195 mb/sec. ...burst
speeds are almost 500 mb/sec.
This can't be right, or at least isn't measuring the same thing as what
would be measured with a single drive. I'm figuring that the max speed of a
single drive to be around 65-75 mb/sec, and Raid 10 should double that
figure (not multiply by 4). The older hard drive tach software worked on
the machine, but provided results that were completely incorrect.
Is any of this meaningful, or won't this software work correctly with
hardware raid. Are there any better benchmark tools that are better?
|
--
Mike Walsh
West Palm Beach, Florida, U.S.A. |
|
| Back to top |
|
|
Rob Turk
Guest
|
| Posted: Sat Dec 23, 2006 9:40 pm Post subject: Re: Raid benchmark question |
|
|
"Jeff" <none@none.com> wrote in message
news:458d3bfe$0$15475$88260bb3@free.teranews.com...
| Quote: |
I'm new to raid and to sas drives, but have been working with parallel
scsi for a few years.
I just put up a new server with 4 Seagate ST373454SS SAS drives - supposed
to be pretty much the fastest things available. They are in a raid 10
setup on an adaptec 4805 hardware raid controller (PCIe).
I tried harddrive tach 3.x on the system, which reports initial read
speeds at the outer portion of the drives/array around 220-240 mb/sec,
dropping eventually to 140 mb/sec, with an average of about 195 mb/sec.
...burst speeds are almost 500 mb/sec.
This can't be right, or at least isn't measuring the same thing as what
would be measured with a single drive. I'm figuring that the max speed of
a single drive to be around 65-75 mb/sec, and Raid 10 should double that
figure (not multiply by 4). The older hard drive tach software worked on
the machine, but provided results that were completely incorrect.
Is any of this meaningful, or won't this software work correctly with
hardware raid. Are there any better benchmark tools that are better?
Thanks
Jeff
|
During writes both drives that are part of a mirror will have to write the
same data, so you'd never see more than double the performance of a single
drive in a 2x2 RAID10 setup. However, during reads each drive that is part
of a mirror can be put to work to handle different read requests. You may
see up to 4 times read performance on the same 2x2 setup. It all depends on
how intelligent your RAID controller spreads the load and if the benchmark
requests data that's spread across all drives.
Rob |
|
| Back to top |
|
|
onetimeonlyposter
GURU
Joined: 06 Nov 2004
Posts: 832
|
| Posted: Sat Dec 23, 2006 10:14 pm Post subject: Re: Raid benchmark question |
|
|
"Mike Walsh" <spamscks@netrox.net> wrote in message
news:458D4D2D.A6E53C28@netrox.net...
| Quote: |
It is theoretically possible to read from a 4 disk RAID 10 array at close
to 4 times the speed of a single drive. Only half the data needs to be
read from each mirrored drive, so they can be read almost as if they were
striped, with each drive reading alternate blocks of data. This depends of
course on the block size and the RAID adapter firmware. This does not
apply to write operations since each drive on the mirror must write all
data.
|
Thanks Mike and Rob. This makes sense. The software will perform two
different types of tests that they call, "long bench - 32 mb zones" and
"quick bench - 8 mb zones." I'm not positive that I know what this means,
but I suspect it is related to what you mention above - if the block size is
optimal for the file size, then it can transfer faster. When I run the quick
bench, it actually runs up to 280 mb/sec but with more variability across
the drive/array (and with about the same average as the long bench test),
which is pretty much about 4x what I would have suspected as the max for a
single drive. So I guess that this software is reporting accurately.
I just flash upgraded the controller's firmware 2 days ago from an adaptec
file with a date about 2 weeks old. They claimed faster read speed (or was
that write speeds) with the new firmware (my real reason for the flash was
due to a warning from Windows 2003 about it being non-certified).
Jeff
--
Posted via a free Usenet account from http://www.teranews.com |
|
| Back to top |
|
|
|
|
| |
Watched Topics
Search
Register
Log in to check your private messages
Profile
Log in
Memberlist
Usergroups