Howdy, Stranger!

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

SER not working properly on dual cpu dedicated server ?

edited June 2014 in Bugs
Hi guys.

Im testing ser on 2 different dedis (i run only ser + cb on servers when testing) to find out whats best deal per buck when it comes to ser performance on server and i noticed something i cant understand.

First dedi - Xeon 1230 (4 cores and 8 threads) 1gbps line, hdd
All here looks ok, i can kill (or nearly kill) cpu using SER. CPU usage in task manager is similar on each core, example screenshot from task manager:

image


Second dedi - Dual Xeon 5650 (12 cores and 24 threads), 1gbps line, ssd
This is very strong dedi, with other seo tools i was getting good results but with ser i cant go above 30% cpu usage - with proxies, without proxies, no matter what i do results (LPM, DPM) are similar or worse than on Xeon 1230 (on both dedis same settings, same lists + only SER + CB running). Ram usage in ser is ~ 1gb so its not problem.

Interesting thing is cpu usage. Its not similar on each core. Some cores have high load, some are nearly not used. Here is screenshot from task manager:

image

Screenshots from Core Temp below - as you can see Processor #0 have way higher usage than Processor #1.

image



So here are my questions to @Sven
1. Ser can benefit from dual cpu dedis (maybe you can update ser for better dual cpu usage)?
2. We should target dedis with more cores or better speed/performance per core?

Someone have same problems as i have?

Ofc something can be wrong with my server, but with other seo tool i was able to nearly kill this cpu ( dual Xeon 5650 ). With Ser its not possible. Here is CPU usage with other seo tool - nearly same load on each core as you can see, and it looks like software is using all cores without problems. 

image



Best Answer

  • SvenSven www.GSA-Online.de
    Accepted Answer

    All I can add to this thread is that SER uses all possible cores even though it looks like one is used more than the other. The main thread (the one where the GUI runs on) might always use more CPU power than the rest. But that can not be changed and is most likely the same for every program you have.

    All other threads are started on the CPU that the OS chooses. I have no influence on this. The OS usually chooses the thread that has the lowest usage at the point a new thread is started.

Answers

  • KaineKaine thebestindexer.com
    edited June 2014
    There are two possibilities: 

    -Hyper Threading not necessarily optimized (and perhaps the instruction sets).

    -Need to increase network bandwidth > 10 gbps (I vote for).

    How many tread / proxy used for your test ?
  • Threads limit - 3000. I got max ~500-600 threads on dual Xeon 5650, 

    Tests (lists filtering - posting to harvested list) were done with :
    - 25 private proxies (not shared), 
    - 600 anonymous public proxies (my own source, uber quality, above 0.25mbps speed per proxy)
    - no proxies

    Proxies used only for posting, and skip proxies for identification. Lits imported to projects from file.

    All tests give same (or nearly same) results in CPU usage / max threads / lpm / dpm.

    Peak bandwidth usage - 100-120 mbps, average 20-40mbps.



  • KaineKaine thebestindexer.com
    edited June 2014
    Can you provide speed test before and in blast ?

    Just for see if connection have the resource. I knows is 1gbps but the total flow and the number of simultaneous connection are two different things.

    500-600 threads with this config does not good ... can't test with verified list ?
  • penumbrapenumbra Antarctica
    Same here, running only gsa ser and usage does not spread correctly on all cores.
  • edited June 2014
    Kaine 
    I do now very important things on this dedi, so i cant provide proof of speedtest while running ser, but i was testing speed with speedtest while using ser and speed was something like 200mbps download, 40-50mbps upload so more than enough.

    Maybe i could test with verified list but list dont have influence on cpu usage and thats main problem from what i see. No matter what list i will use speed should be way better on such dedi.

    I really appreciate that you are trying to help me.

    Thanks for info. Looks like you have same problems as i so its not releated to my server. Hope @Sven can take a look at this.

    JudderMan  @Gooner You guys dont have such problems on dual six core dedis?
  • KaineKaine thebestindexer.com
    edited July 2014
    Hi @Sven,

    Have you make an optimisation on version 8.71 for cpu ?

    I see my cpu can climb higher :)

  • SvenSven www.GSA-Online.de
    not really no.
Sign In or Register to comment.