• State
  • Locked Locked
  • Replies 3 replies
  • Subscribers 99 subscribers
  • Views 347 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Network Performance Issues

Greg Sulfid
Intellectual 610 points

Other Parts Discussed in Thread: AM3517

Hello,

first of all I encounter a problem with the network throughput. When I transfer a file with FTP, I get a maximum of 600 kbytes/s.
When I run CETK socket performance test, 1500 kbyte/s is the maximum and CPU level is at 100%:

 366510 PID:934000a TID:818000e      Send Packet  Bytes Sent   Send Rate    CPU Util
 366512 PID:934000a TID:818000e      (Bytes)      (Bytes)      (Kbps)       (%)
 366514 PID:934000a TID:818000e      -----------  -----------  -----------  --------
 373445 PID:934000a TID:818000e    *          16       160000       439.11     97.91
 377759 PID:934000a TID:818000e    *          32       320000       653.73    100.00
 382467 PID:934000a TID:818000e    *          64       640000      1195.70    100.00
 387512 PID:934000a TID:818000e    *         128      1280000      2187.10    100.00
 392575 PID:934000a TID:818000e    *         256      2560000      4235.78    100.00
 398654 PID:934000a TID:818000e    *         512      5120000      7043.85    100.00
 406731 PID:934000a TID:818000e    *        1024     10240000     10368.31    100.00
 465815 PID:934000a TID:818000e    *        4096     40960000     15277.88     94.88

Receiving data is at a maximum of 1000 kbytes/s.

The same behavior in the release version without VMINI and also on an USB network adapter (LAN9514).

I called "BSP Information" to get the CPU speed, but it only shows zero. When I set a breakpoint at OALIoCtlHalGetCpuSpeed, it never gets there?! When I rely on eboot, it should run at 600MHz and even with 125MHz the performance shouldn't be so slow?
Did not look deeper at the output message "****Profiler Build****", but if I'm not wrong, there should be no profiling which could cause such a latency (function call profiling or something)?

Someone any idea?

  • 0
    TI__Prodigy 250 points

    Greg,

    Please provide the following information so I can get a better idea of your setup:

    Platform type/model (custom board??) being used:

    Operating System on your platform (CE6/CE7):

    Which code build:

    Is EVM connected directly to a PC or is it going through a switch/hub (make/model):

    Operating system on the PC (WinXP, Win7, etc):

    Are you running the CETK performance test from within Platform Builder?

    Regards,

    Ken

  • 0 in reply to kengler01
    Intellectual 610 points

    Hello Ken,

    sorry, my mind was so fixed at the problem, that I forgot the basic informations.

    Its a custom board with the AM3517 and we use your latest CE6 BSP 1.2.0.

    We tried the connection over a hub (3Com Office Connect Dual Speed Hub) and a switch (Asus GX1180N) and also with different PC and OS (WinXP, Win7).

    Don't know, what you mean exactly with running the CETK from within Platform Builder - for the release version I used the manual connection and nothing over KITL.

    For me its a little bit strange, that the performance on the internal EMAC is as bad as on the USB LAN chip SMSC9514 - and also the high CPU usage.

    Thanks ahead!

  • 0 in reply to Greg Sulfid
    TI__Prodigy 250 points

    Greg,

    Here are the comparative results between the various platforms and CE6/CE7 (link below).  Since we are still in the process of testing CE7, the Ethernet performance results are listed as TBD on a couple of the platforms.

    In all cases however, much higher results are being seen than those you are seeing. Also, be aware we are reporting network performance, as Megabits/Sec which is how the CETK tests report them.

    http://processors.wiki.ti.com/index.php/WinCE_Comparative_Benchmarks

    Next week, I'll try to take a closer look at things since I'll have more time to do some additional testing.

    Regards,

    Ken