Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


Way to determine if the node CPU is being maxed?
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

Way to determine if the node CPU is being maxed?

cr08cr08 Member
edited February 2012 in General

Long story made short: I am trying to determine if there is any way from inside a VPS instance (KVM in this case) to determine if the node CPU is bottlenecking while trying to run a given application? Since usually the VPS (in my limited experience?) tends to see what the fully allotted CPU is capable of doing and bases current usage inside the VPS off that so in many cases it will never reach 100% and could be maxing out like 50% tops when fighting with other VPS' on that node.

Reason I say this is I am currently trying to debug a strange issue with an application on my Windows based KVM that I am widdling down possible explanations to lack of CPU at this point. Packet loss and available bandwidth have been since thoroughly eliminated (Continuously run traces to numerous destinations read very, very low loss if not 0% and there's an abundance of bandwidth for what this application needs in just testing).

Comments

  • Create an application doing nothing bot looping and incrementing a counter. Make that counter externally accessible. Run this application in the background, with lowest priority.

    This will practicly use all the idle CPU time. By measuring how fast the counter increments, you can measure how much CPU time this application gets.

    But don't just leave this application running for too long, otherwise it would be considered CPU abuse.

  • @rds100 said: But don't just leave this application running for too long, otherwise it would be considered CPU abuse.

    If the provider is dumb, yes lol. Because you already reniced the process :P

  • I'd do a variation on rds100's idea. Make something that uses on the order of a second of cputime. Schedule it to run every five minutes or so and record the actual time. I would not expect an oversold machine to have even load during a 24 hour day so you should see variations.

  • You could always ask the host to send you the stats on the node, If one of my clients wanted to know the CPU usage over all I would have no issues sharing that.

    Thanked by 1Amfy
  • Sorry for the response delay. Haven't yet had the opportunity to submit a ticket or do my own testing. But here is the direct issue that caused this questio to come up: I currently have a 512MB KVM with BuyVM running Windows 2003. I am in the process of testing some audio streaming to see if the VM would handle it. Specifically it also does live encoding and not just serving off the same box. Start up the process, tune in (from any player/device) and after a minute or two there's constant buffering issues. No matter the codec or bitrate. Bypass encoding on the VM and stream from Winamp locally up to the Icecast/Shoutcast servers running on the VM and tune in from another client. Not a single buffer. Bandwidth should not be an issue. Speed tests I see no less than 10mbit on the upload from the VM and this is like 1 listener at anywhere between 32kbps and 128kbps. CPU, in theory, should not be an issue either but I am still out on debate. Given benchmarks always hit 50% average every run (and this might simply be because the benchmarking program is single threaded on 2 cores, didn't check specific core usage) and while the encoding application in question is going I see between 5-12% usage tops, I am somewhat leery on looking at this as a possible cause.

    Any smart people got any possible theories I can test? :) I have one real out there theory which is possibly the clock as one benchmarking app I tried had complained about clock inaccuracies. Related to that, I am running the Win32 version of NTP to keep the (time) clock in sync.

  • FranciscoFrancisco Top Host, Host Rep, Veteran

    It's probably you just need the virtulization calls the L5420's are missing.

    The L5420's are good for heavy work loads but suck for threading performance in virtulization. If you could give us a chance to get KVM08 on an E3 and check then that'd be nice :)

    08 is running at like 5 loads on the HN, but I think you're just feeling the emulation of all of the calls that the CPU's can't do in hardware.

    Francisco

  • Sure thing! I'll see how it fares once the E3's in and report back. :)

  • FranciscoFrancisco Top Host, Host Rep, Veteran

    KVM04 got some sweet franlove last night, i'll see if I can get KVM08 for tomorrow even.

    Francisco

  • Any updates? Not in a huge hurry, but just curious. Kinda hard to see CPU changes from inside Windows so mostly in the dark. :)

Sign In or Register to comment.