Slow response from disk if idle

[sullivang]

System:
Dell Latitude E6500 laptop, P8600 Core 2 Duo 2.4GHz, internal Intel ICH9M-E
SATA storage interface, Windows XP SP3, Samsung HM251JJ 250GB SATA HDD, NTFS
file system.
Drive power down is disabled in power management profiles.

Problem:
Using the IOMeter disk excerciser utility, I am observing rather slow io
response times (~360ms), if the drive has been idle for more than
approximately 5 seconds. If the drive is accessed more frequently than every
5 seconds, every io completes rapidly. (response time in same ballpark as
drive access time specification)

Similar behaviour can be reproduced on an external USB drive (HP SimpleStore
1TB), however the idle timeout with this drive is 8 seconds, and the initial
response time from idle is over 400ms.

On a third drive, being a very old IDE drive in an USB enclosure, the
problem does not occur at all. I've gone as high as 20 seconds between io's,
and the response time is always rapid.

The above behaviour is observed when using the "Always On" power management
profile. I have been using this profile because it disables processor speed
throttling. If I switch profiles to either "Minimal Power Management", or
"Maximum Performance (Dell)", the response time from idle is greatly
improved. For the internal drive, the typical response time is reduced to
40ms, and for the external HP drive, 60ms. (I haven’t tried the old IDE drive
with these profiles yet)

I thought I had found the solution, and all I would have to do now is
disable processor speed throttling. Unfortunately, when I do this (using
POWERCFG.EXE), the slow response times return! I am very puzzled by this.

The reason I am doing this testing is that a real-time audio application
that streams from disk is failing, and I believe this issue is the root
cause.
The application works if I generate tiny bit of background disk activity. It
also works fine from the very old IDE drive, although with limited
performance.
I have not yet verified that the specific problem I am having is fixed by
using “Minimal Power Managementâ€, or “Maximum Performance (Dell)â€). From
past experience, though, processor throttling doesn’t work very well in
general for this kind of application, and I should not have to use it.

Any insight much appreciated.

Regards,
Greg.

 

[sullivang]

Andrew,
Yes, the ICH9M-E uses the Intel Matrix Storage Manager software. I have
contacted Intel, and they do not think the controller could cause this
behaviour.
In any case, as I said, I can reproduce the same behaviour on an external
USB drive, which presumably has nothing to do with the ICH9M-E whatsoever.

I have no reason to suspect that the drives are "failing". I am seeking an
explanation of the behaviour I am observing. I am especially interested to
know why, when I disable processor speed throttling in the power management
profiles "Minimal Power Management", and "Maximum Performance (Dell)", the
slow io response times occur again. It appears to me that disabling
throttling also changes some other setting, which I have no obvious control
over.

I will nevertheless investigate those two downloads - thankyou.

Greg.

 

[sullivang]

The two profiles for which the io response from idle is fairly rapid do NOT
actually work with my audio application - it still fails in the same manner.

With the app loaded, I decided to run the IOMeter test again. As I
suspected, the response time was slow again. For whatever reason, having the
app actually running has effected the io response time.

I think the main issue here is power saving states of the disk drive. I
have learnt that some drives do have multiple power savings states, such as
turning off the heads, or even parking the heads. Being a laptop drive, it's
hardly surprising that it's doing this.

Generating the background activity is an easy workaround for me for the time
being. I've just been using the app with a background random read of 512
bytes every four seconds (i.e a tad under the idle timeout), and it works
fine.

I think the subtle changes in behaviour with the various power savings
profiles are going to be very difficult to diagnose/explain.

Greg.