Sign in with
Sign up | Sign in
Your question
Solved

Slow initial response time from drives.

Last response: in Storage
Share
December 15, 2009 8:50:20 PM

I have a problem that *appears* to be related to idle behaviour of disk drives:

System: Dell E6500 laptop running XP SP3, processor locked on maximum frequency and drive power down disabled.
I am using NTFS for the io testing - NOT raw partitions.

Using the IOMeter utility, I am noticing that the io response time of hard disks is very large (hundreds of milliseconds), if the drive has just been idle. I.e - the very first io to the drive is very slow, and from that point on it is fine. I am doing random reads of 512 bytes, with a queue depth of 1. (no queuing)

If the time interval between io's is sufficiently high, then EVERY io takes a long time.

E.g: internal Samsung HM251JJ connected to Intel ICH9M-E RAID controller: if it is not accessed for approximately 5 seconds, the response time is about 360ms. If it is accessed more frequently than that, the access time is significantly lower. (much closer to specification)

Another example: an externally attached USB 2.0 drive (HP SimpleSave 1TB), the idle timeout is about 8 seconds, with similarly large response times if accessed at this interval or greater. I can reproduce the same behaviour when this drive is connected to a desktop computer.

Samsung say the seek from idle of their drive is only 5.6ms, and they have recommended I contact the manufacturer of the computer. I have not yet contacted HP. (it is the Samsung drive that I am most interested in)

Trying a third drive, which is a very old ATA drive (in a USB enclosure) - this one does NOT suffer from this problem! I went as high as 20 seconds between each io, and the response time remained very low.

It's looking like it's the drives themselves at this stage. Maybe power saving, or preventative anti-shock head parking, or something?

The reason I am doing this testing is that I have a problem with a real-time audio application, and it appears that this idle behaviour is the root cause. The same applicaiton works properly when I use the very old IDE drive. (although overall throughput is not very good, as expected)

As a matter of courtesy, note that I have cross-posted this on the Anandtech forums, and I may post elsewhere too.

Any help will be much appreciated!

Greg.
a b G Storage
December 16, 2009 6:08:36 AM

I would say power saving. How often does this happen? Every time you use the disk, or after your computer hasn't read/wrote anything to/from the disk for a relatively long time?
m
0
l
December 16, 2009 7:08:20 AM

For the Samsung, if it is idle for 5 seconds or more, the initial io will be slow. Yes - every single time it is accessed. If I run IOMeter with an inter-io delay of, say, 7 seconds, both the average response time and the maximum response time will be about 360ms.

For the HP SimpleStore, it's the same, except that it has a slightly longer idle timeout period. (8 seconds instead of 5 seconds)

I agree that it does appear to be power saving, however Samsung aren't saying this yet.

Greg.
m
0
l
Related resources
December 16, 2009 7:12:04 AM

(not that it matters, but for the HP, the response time from idle is actually higher - over 400ms)

Greg.
m
0
l
a c 415 G Storage
December 16, 2009 3:52:13 PM

The Samsung HM251JJ is designed for laptops, so it's doubtless designed to power down after a brief period of inactivity. What you're seeing would be perfectly normal for this kind of drive.

The only way to avoid the issue is to use performance drives that don't spin down when idle.
m
0
l
December 16, 2009 4:29:30 PM

Again - that's what I thought too, and I agree that you're probably right.
However, Samsung said that the initial seek, from idle, takes only 5.6ms.

The drive is not SPINNING down - that is disabled. (360ms is way too fast for a spin-up, anyway, I assume)

I am attempting to have this question escalated to Samsung's engineering department.

Greg.
m
0
l
December 16, 2009 4:32:34 PM

Just btw, the audio application also works with the Samsung drive, if I use IOMeter to generate a tiny bit of background activity on the drive. (e.g a read every half a second, although it looks like it would also work with a read every three or four seconds, from the results of my testing)

Greg.
m
0
l

Best solution

a b G Storage
December 16, 2009 4:48:08 PM

sullivang said:
Again - that's what I thought too, and I agree that you're probably right.
However, Samsung said that the initial seek, from idle, takes only 5.6ms.

The drive is not SPINNING down - that is disabled. (360ms is way too fast for a spin-up, anyway, I assume)

I am attempting to have this question escalated to Samsung's engineering department.

Greg.

There are quite a lot of power saving states on a laptop drive, ranging from fairly mild to fairly aggressive. If you've been idle for several seconds, it's entirely possible that the drive has parked the heads and turned off the controller, which would account for the extra latency.
Share
December 16, 2009 5:09:57 PM

Thanks Cjl.
I've located a document which mentions these various power saving states on the Hitachi Deskstar drives: http://tinyurl.com/b52nwf

Greg.
m
0
l
December 19, 2009 12:41:12 AM

I've discovered that switching power management profiles greatly improves the initial response time. For example, "Minimal Power Management" and "Maximum Performance (Dell)" both produce rapid response times from idle. This applies both for the internal Samsung and the external USB drive!

I had been using "Always On", for the reason that this profile disables processor speed throttling. These other two profiles enable throttling, which is not appropriate for my application.

Very oddly, if I change the processor throttling to NONE using POWERCFG.EXE, the response time reverts back to what it was with the "Always On" profile.

Any insight would be much appreciated!

Greg.
m
0
l
December 19, 2009 6:17:00 AM

After further testing, I think the change in behaviour when changing the power management profile is something of a red herring. I'm now quite confident that the main problem is just the idle behaviour of the disk drive. The only thing remaining now, for me, is to get confirmation from Samsung that the drive is performing to specifications. I'll be extremely surprised if it is not.

Thanks everyone!

Greg.
m
0
l
a b G Storage
December 19, 2009 7:11:18 AM

I've talked to a friend I know at Seagate, and they say that assuming Samsung drives have similar idle behavior to Seagate ones, it should be completely normal.
m
0
l
April 9, 2011 8:56:27 AM

cjl said:
I've talked to a friend I know at Seagate, and they say that assuming Samsung drives have similar idle behavior to Seagate ones, it should be completely normal.



I've built an application and the slow io time is playing havoc with its usability. Same exact problem Sullivang is having. So, I switched to a ssd and completely got rid of the standard hard drive. I AM STILL HAVING THE SAME PROBLEM. Is this a driver issue with the processor, maybe a cache issue? I am at a loss. Will post back here if i solve the problem. In the mean time if anyone knows how to fix this it would be super helpful.
m
0
l
April 9, 2011 10:27:11 AM

elin99 said:
I've built an application and the slow io time is playing havoc with its usability. Same exact problem Sullivang is having. So, I switched to a ssd and completely got rid of the standard hard drive. I AM STILL HAVING THE SAME PROBLEM. Is this a driver issue with the processor, maybe a cache issue? I am at a loss. Will post back here if i solve the problem. In the mean time if anyone knows how to fix this it would be super helpful.


FWIW, Drive Power Manager solved the problem for me: http://www.drivepowermanager.com

I suspect you have a different problem though, given that the SSD exhibits the same symptoms.

Greg.

m
0
l
!