Jump to content

Vista Low Latency Checker .......this COULD be the KEY !


benifin

Recommended Posts

Hi nuhi

Again, you are a GENIUS.

Check this out:- http://www.thesycon.de/dpclat/dpclat.exe - very small 200k d/l - standalone test application

Its a Windows [ XP32/XP64/Vista ] Low Latency checker.

Its brilliant.

The lower the values and less spikes the better.

On my XP32 rig and my XP64 rig, it settles down around 7us to 15us and never spikes above 35us to 40us.

With VLite .95 32 bit - fully stripped and most services disabled - it sits around 35us to 75us and peaks every 30/40 seconds at +1000us - thats makes Vista 8 - 10 worse than XP32 or XP64.

Clearly, there is a lot of "stuff" [ crap / drm / polling ] going on in Vista that is not there in XP32 or XP64.

Maybe this tool will help you identify where all the cpu cycles and memory useage is, and how to "dis-integrate" them.

I hope this tool can help you further "cleanse" Vista and identify areas of even further improvement so we will hopefully soon see Vista operating at the same or better speed, efficiency and latecny as a currenlty tweaked XP32 r XP64 system.

Again, you are the man !!

Ben

Edited by benifin
Link to comment
Share on other sites


Excellent tool!

Thanks for the heads up.

edit: I have a question. What does it actually measures? I run few programs, even start nlitement and it doesn't change the usual peaks...

Link to comment
Share on other sites

I am quite confused about the data provided. Are you drawing a comparison in Latency between XP and Vista vLite'd? I would expect a compsarison between a Windows version and a *Lite'd Windows version.

Link to comment
Share on other sites

Hi nuhi

Its essentially a measure of core and sub-core "noise" / "activity" which "gets in the way" of an uninterupted data flow between any external hardware input or controller device and the core of the operating system.

As such, lower readings and less spikes are critical.

The figures I quoted compared a stock XP32 clean install and a stock XP64 clean install to a Vlited [ .95 fully stripped Business Version ] install.

If you run it in a clean full non-Vlite'd Vista install [ Business ] the results are pretty much the same as the Vlited [ .95 fully stripped Business Version ].

This suggests [proves to me] that there is some major memory and/or cpu cycling and/or driver /usb polling etc... going on at very-low [core] level which currenlty makes so sluggish [ compared to XP ].

I reckon if nuhi can monitor future releases and get the latency down to XP32 or XP64 levels using this tool, the "holy grail" will have been achieved - ie: what Vista SHOULD have been - faster, nimbler, more-reponsive, and more software and hardware tranparent than XP32 or XP64.

Ben

Link to comment
Share on other sites

i did suspect the ram usage to be more or less the same with vlite and non vlite earlier. but not sure since ive never removed superfetch as it benefits what i work with.

latency on my xpsp2 with latest updates is around 70 with absolute max at 500

however you have a good point worth investigating.

at that low a level however it could be very complicated to reduce ram usage.

but with nuhi genius one can never write off the possibilty.

go guys hope the next version works on reducing ram footprint just like the iso has been drastically reduced in previous versions.!!

Edited by pallavsuri
Link to comment
Share on other sites

Bilar Crais - spot on.

pallavsuri - I've no doubt that reducing the RAM footprint to around 100 - 150 meg on a clean boot up will simultaneously address the latency issue which is the direct cause of the sheer "sluggishness" that Vista has.

I also have absolutly no doubt that as nuhi keeps "dis-integrating" all the Vista components and services and "features", that the "tip-over" point will be crossed - ie: the point where Vlite will be as snappy as XP32 and XP64 on a same-spec'd machine.

I am certain that the RAM footprint and the system latecny are interlinked - address one and the otther will automatically come .... and them ....bam...we'll have the Vlite we all want.

There is no doubt that nuhi is not only the best person to achieve this, but that he will - and I reckon he's a lot closer than perhaps even he thinks.

Ben

Link to comment
Share on other sites

In my opinion, removing components might be good if done properly. When components are reported missing in Event Log, then this most likely increases the latency, where as when those componnts are present and started (in case of a service), then latency is not influenced or at least not as much.

Maybe if there is some way to remove the (performance) monitoring from Vista (not sure if this is a component), then the latency might be reduced. Under XP there was an option to disable Performace Timers(?), not sure if this can be done for Vista.

I can understand some people find it useful to see Reliabilty and performance monitor, but I rarely use them. Only for fun from time to time and I add not value to it. I know that if I am overclocking and I have system crash that my system is not that reliable, but if I alter system settings like voltage, then stability is increased, but this not reflected in Reliabiluty monitor. In other words: drop these montioring and logging services if possible.

I do want to give a thumbs up to the topic starer: a lot of people look at memory usage and the amount of services started, but this is also an important aspect. I do not share is opinion that this is the holy grail, it is really important but a lot of aspects are important as well. there is no 1 holy grail.

Link to comment
Share on other sites

Jeronimo, indeed, performance counters can be disabled, or at least removed, it's just a reg entry in the services I think. I delete those performance folders in all of my services and then the performance counter configurator doesn't display any of them.

About the Event Log...that one might get tricky as XP needed some hacking to allow the removal and I would like to postpone hacking Vista (if even possible on 64bit version) until it's the only big thing left. But then again maybe it removes just fine, we'll see in one of the next versions.

Just to repeat what I said...that latency checker doesn't change the values whatever I run...then what is it actually measuring is the question, maybe it doesn't have anything to do with the processes in the background, maybe it all depends on the CPU driver used and ACPI pooling.

Link to comment
Share on other sites

Just to repeat what I said...that latency checker doesn't change the values whatever I run...then what is it actually measuring is the question, maybe it doesn't have anything to do with the processes in the background, maybe it all depends on the CPU driver used and ACPI pooling.

Same here, I can't figure it out... mine constantly stays at 1003 +/- 10 us, and no matter what I do it doesn't change. Spikes occur once at ~30 secs with an absolute maximum at 1460 us.

I have web server, application server, dual gigabit LAN transferring data, visual studio, VMware 6, all running and no influence.

So what am I measuring???

Link to comment
Share on other sites

@benefin - dis integrating components will reduce the ram footprint. Additionally it may not reduce the latency because vista keeps checking for the components even when they are not present probably more so when they are not present adding to the i/o clutter and 'sluggishness'. example scm error in event log.

ideal scenario would be where components removed are not checked for by the OS.

@nuhi - i'd request you to try out the hacking bit for the next version and see the latencies outside vmware - i think thats the direction where the wind is blowing. might as well set your sails in the direction! who knows we might not need any component removal to reduce the latency and consequently ram footprint right away instead of waiting 8-10 months! performance boost comes now and component removal can take its own sweet time in later versions. no hurry for more component removal since you have added the split functionality anyway. :D

Also can the poster of the topic translate what the coder says about this program dpclat?

thanks

goodluck

Link to comment
Share on other sites

Where can I find this Performance Counters Configuration tool. I had it in the past, but I can not find it anymore.

I am at 92us. I have a C2D E6400 @ 3200MHz, but I have EIST enabled so it runs at 2.4GHz (6x400) idle. When disabling it and running at 3.2GHz (8x400), DPC whows consistent around 86us.

Disabling my Avast! virusscanner provides no decrease in latency whatsoever.

My absolute maximum is at 23us, quite high it seems.

However I this does not say anything to me now, beside the fact that with latency less is more (better). I have not read the DPC guide and I intend to do this first before I can comprehend the full meaning and useability of the tool.

Edited by Jeronimo
Link to comment
Share on other sites

when they are not present probably more so when they are not present adding to the i/o clutter and 'sluggishness'.

Isn't this speculation?

Standard development practice is to check things on a decreasing scale. For example you might check something every second for a minute then you might check every 10 seconds and so on. Of course I am also speculating but based on real development experience.

You could be right but for the wrong reason. If something is not present and is 'supposed' to be then this would get logged. Obviously if the device was present there would be no need to log the fact although as mentioned Vista does have some pretty heavy logging. I find it strange that this stuff is all on by default. Surely for the end-user it's not necessary. Companies may want to turn it on so enable their IT departments to more easily diagnose issues but I can't see Dell having support that goes into someones home computer to diagnose issues. Correct me if I'm wrong.

Edited by Dobby
Link to comment
Share on other sites

I tried disabling all performance counters, but no impact on latency. First reboot gave me a blue screen, 2nd boot worked and checked in registry that counters where disabed which was the case.

I also had no spikes when my pc was idle, so no comparison for me if this was reduced.

Edited by Jeronimo
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...