Learn how easy it is to sync an existing GitHub or Google Code repo to a SourceForge project! See Demo

Close

Performance between the tcpcm and ethcm

cdlike74
2012-09-17
2012-09-28
  • cdlike74
    cdlike74
    2012-09-17

    Hi all
    I setup two nodes ( one is RedHat 5.0 x86 host - server , the other is the
    powerpc target - client ) with ethernet cm and tcp cm, try to do a evaluation
    of performance between the two diffient connection . In my mind I think the
    ethernet will be more efficient thant the TCP , but the test result suprise
    me. following test based on the linx_bmark provided by linx library 2.5.1 , .

    1) ethernet connection
    -bash-3.2# ./linx_bmark -p target_89/server -n100 -c2 -m50 -b1000

    %% Running LINX burst bmark without linx api %%
    Test parameters:
    Burst count : 1000
    Hunt path : target_89/server
    Loop count : 100
    Running subtest 1 of 1 with message size : 50 bytes
    Result:
    Average : 0.0 us
    Min : 157964 us
    Max : 1298878 us
    Diff : 1140914 us
    CPU load : 0.00%
    user : 0%
    system : 0%
    nice : 0%
    idle : 48%
    iowait : 0%
    irq : 3%
    softirq : 48%
    Com test 4 completed
    -bash-3.2#

    TCP connection :
    -bash-3.2# ./linx_bmark -p link_89/server -n100 -c2 -m50 -b1000

    %% Running LINX burst bmark without linx api %%
    Test parameters:
    Burst count : 1000
    Hunt path : link_89/server
    Loop count : 100
    Running subtest 1 of 1 with message size : 50 bytes
    Result:
    Average : 0.0 us
    Min : 40467 us
    Max : 53376 us
    Diff : 12909 us
    CPU load : 0.00%
    user : 0%
    system : 100%
    nice : 0%
    idle : 0%
    iowait : 0%
    irq : 0%
    softirq : 0%
    Com test 4 completed

    does anybody do same test like this and got same result ? why does the TCP
    connction get more efficient thant ETH connction ?

     
  • Hi,

    I wrote my benchmark application by my own. I test LINX for Linux 2.5.1 using
    10k packets for each packet size (64B, 1024B ... 65536B). I also got TCP/IP
    solution much faster in inter-node operability than when using ETH. For
    example, 4096B packets where pushed to ETH buffer in around 61 us in average.
    Same size of packets using TCP/IP was pushed out in around 3,24 us in average.
    However this is the matter of throughput only I believe. The round-trip delay
    scenrio shows that ETH is a little bit faster than TCP/IP.