How much CPU usage is required to slow system performance down?

[Simfish]

So I know that my CPU is now slowing down, even though Process Explorer reports system usage as varying between 30-70%, far short of 100% system usage. What is the most reliable way to measure system usage before the system starts to slow down?

 

[jim mcnamara]

What impacts 'perceived' performance:

If you are running something, like an editor, and you hit a key and there is a wait for the screen to display it, that's a perceived performance hit. The performance may only be degraded by .5 second.

On the other hand, if you're printing a file and the same thing happens to the spooler (the process that prints for you like winspool) you would never notice.

So, two things that slow down your perceived performance are:

I/O bound - your process is competing for the same files as another process
or you have to wait for another process to complete it's I/O request.

CPU bound - you have to wait for the other guy to release the cpu to get
your share.

I/O bound is probably what you are experiencing. You can be I/O bound and have low cpu usage. I/O bound is related to I/O queue request length.
Like being in a line 5 people behind the head of the ATM line. When you get your money, you go to the end of the line to wait to get more money. The longer the line the longer the wait.

There is no 'most reliable' way. User perception varies depending on what they expect. If a user starts a job that last 30 minutes and it completes in 33 minutes they probably won't care because they went to get lunch anyway.

If the user is entering data and he has a perceptible delay in screen update after a keystroke -say .2 seconds, he'll complain. If the overnight batch run grinds on until 10:00am everybody will scream loudly, including your boss.
That is a huge performance hit.

Performance is in the eye of the beholder. If you want a metric look up Dhrystone. That is, performance suites that measure raw power. Real performance is what the user sees/(does not see) under normal (whatever that is)load.

 

[russ_watters]

Hard drive usage is usually a bigger factor, but your processor should not have between 30 and 70% usage all the time. What process is using all that power?

 

[chroot]

In modern pre-emptive multitasking operating systems, it's often the case that CPU utilization has nothing at all to do with a user's perception of speed. As long as the processes and threads which interact with the user are given a higher priority than batch processing tasks in the background, the user can comfortably use a computer even with most of its processor time spent doing other things.

Russ is right that the culprit is often not the processor utilization, but the I/O bandwidth bottlenecks to and from a hard drive. This is particularly true if you have a small amount of RAM, and the operating system has to temporarily exile portions of actively running programs to disk.

So, look at your disk access light. Look at the amount of RAM available. Look at the number of page faults per second (each page fault means the processor needs a page of memory that has been sent to disk and is no longer present in RAM).

You may find that disk-intensive tasks like anti-virus software, search-indexing processes, etc. will have the biggest impact. Look for those first, and disable or reconfigure them.

- Warren

 

[Simfish]

Thanks for the responses!

Ah - yes - when I'm merely working on a task on firefox, the CPU usage usually alternates between 10-50% (10% seems to be the base level).

The lag times still show up even when I have 30-40% of the physical memory free. (50% is the maximum amount of memory that is ever free - is this perhaps due to Vista's need for physical memory?)

0-100 faults per second. The page file is pretty consistent: 1300/2000 M, even when launching applications.

EDIT: I just tried booting up Internet Explorer under such conditions - there was a lag time of approx 25 seconds until I could finally use it.

==
What explains, say, the lag times in loading up firefox or Windows Media Player? (when there are still decent amounts of RAM and CPU processing power left?) The number of page faults don't seem to significantly increase whenever I boot up Firefox or Windows Media Player.

==
Also, there are a number of background system processes - each of which use up 22 MB of RAM (out of my 900 MB of total RAM). Will they produce a perceptible impact on performance when 30% of my RAM is still free?

 

[russ_watters]

As warren asked - what does your hard drive light look like? Maybe you need a defrag...?

I have 50 background processes that are currently doing roughly nothing and they have no effect whatsoever on the perception of speed on my computer.

 

[rcgldr]

If the lag times are significant, you may have a virus on your computer, which is consuming CPU bandwith, or just causing delays.

If you have Windows Media Player 11, it has an issue shutting down if you terminate it before reaching the end of a video. Although the app will shutdown quickly, the process (viewable from Task Manager) won't. Sometimes it takes about 10 seconds for the process to terminate before you can re-launch it to view another video. I'm not sure if WMP version 10 has this same problem.

 

[Simfish]

Hard drive is 1% fragmented and Hard Disk Sentinel records it at 100% health and performance.

Which is really kind of weird... it actually had such lag times since I bought it - though they're noticeably worse than before, hmm...

 

[russ_watters]

...3rd try: what does your hard drive light look like? Is your hard drive being accessed constantly?

 

[Simfish]

hm yes - it is constantly blinking even when idle.Is there a way to check the the "busy-ness" of the hard drive on a numerical scale relative to maximum I/O performance of the hard drive?

There's the resource monitor of Windows Task Manager - do the response times add up? Largest three are searchindexer.exe, one at approx 1209 ms, and two at approx 600 ms

Are response times an indication of I/O bottleneck?

 

[chroot]

A little blinking every now and then is probably due to tasks like search indexing, spyware scans, etc. This is entirely normal.

What we were concerned about is protracted, long-term periods of very heavy disk activity.

If you're running a Windows operating system, it sounds like you might just be succumbing to registry rot. (The registry gets filled with garbage by programs you've long since removed, and the computer gets slower, and slower, and slower...) You might consider simply re-formatting and re-installing your operating system.

Alternatively, you could install a fresh copy of your OS on an external hard drive, and boot off that just for verification. Hopefully you'd notice a major improvement in responsiveness.

- Warren

 

[Simfish]

Ah, I see. Yeah - that's a good idea - I just requested a copy of Ubuntu and will probably try it out on my external HD when it comes.

The number of hard faults per minute currently resulting from the application instance.

Note
A hard fault (also known as a page fault) occurs when the page of the referenced address is no longer in physical memory and has been swapped out or is available from a backing file on disk. It is not an error. However, a high number of hard faults may explain the slow response time of an application if it must continually read data back from disk rather than from physical memory.

Hm - I just found out about this. I tried booting up IE, which peaked out at around 500 hard faults/second (as it was loading). Firefox gets 1000 hard faults/second when it is busy. Perhaps it's hard faults/second that is the culprit.

 

[chroot]

Page faults are entirely normal while an application is loading. You only need to be concerned when the application has tons of page faults during normal use.

- Warren

 

[Simfish]

Ah yes - Firefox actually did have 1000 page faults during normal use - though when loading a new webpage. Anyhow I was just looking for an explanation for the sometimes slow-performance of Firefox during normal use - and it looks like it's page faults.

 

[chroot]

Again, a few hundred page faults (a couple of megabytes of memory) are not unusual. What we were concerned about, again, is long, protracted, and unyielding periods of swapping back and forth to disk. This does not appear to be happening.

Nothing you're describing is unusual.

I'm beginning to think your concerns about responsiveness are due to deficiencies specific to Windows itself, and it probably needs to be re-installed.

- Warren

 

[Simfish]

Oh, I see. Thanks for the responses everyone! Yeah, I should think about trying that out.

 

[russ_watters]

A little blinking every now and then is probably due to tasks like search indexing, spyware scans, etc. This is entirely normal.

To quantify that, my computer, when otherwise idle, accesses the hard drive just a few percent of the time. Ie, every few seconds, it accesses the hard drive for a small fraction of a second. A computer running Windows 2000 or earlier, without too much extra stuff, will never access the hard drive when otherwise idle.

For CPU usage, I haven'd done anything but access PF in the past 5 minutes and in that time have not seen above 10% CPU usage.

It really shouldn't be necessary, but as warren implies, I rarely go more than a year or two without a clean Windows install. And if I buy a new machine, what comes pre-loaded on it gives me the heebie jeebies. Windows is just not very good at taking out it's own trash and eventually gathers crap that can't be removed any other way than a clean install.

 

[Simfish]

I see. Do I have to format my HD for a clean install of Windows?

 

[chroot]

I would recommend it, yes.

- Warren

 

[Simfish]

Heh, interesting. I uninstalled all of the bloatware pre-installed with my PC and deleted my Startup folder and now everything's much faster. :) Ugh - I should have realized that earlier.