Strange performance issue

Discussion in 'Networking Issues' started by Trollkarlen, Aug 25, 2005.

  1. Trollkarlen

    Trollkarlen Network Guru Member


    I recently bought a Linksys WRT54GS v1.1. The purpose was to replace my D-Link DI-604 since the D-Link is lacking wireless capabilities. However, since I can't get the throughput of the Linksys to match D-Link I'm currently using the Linksys as gateway and accesspoint only.

    Does anyone know what kind of throughput you can expect between WAN <-> LAN (wireless excluded) on a WRT54GS v1.1? For some reason I can't get more than roughly 5 MBit WAN -> LAN and about 7 MBit upload (LAN -> WAN), independently of which firmware I use. However, if I instead use the D-Link DI-604 as my primary router and only use the Linksys as a gateway/accesspoint I get download speeds of 8 MBit or more.

    These are the two configurations I've tried:
    Now, here's the weird part - Config. 2 above gives me the highest throughput (8+ Mbit for downloads). Could it really be that the processing power for the GS is so bad that when it has to do all routing and stuff it can't handle higher throughput than the 5.5 MBit I'm seeing?

    I have tried several different firmware releases (Linksys 4.70.6, Alchemy 1.0, HyperWRT and DD-wrt) and I always get the same result.

    Any help in solving this is very much appreciated!
  2. 4Access

    4Access Network Guru Member

    I just did a quick test and got about 13.5Mbps (1.7MB/s) on my WRT54GS

    Differences between our tests:
    - I used a WRT54GS v2 (but they have the same hardware so it shouldn't matter)
    - I had a 100Mbps Full-Duplex computer connected to the WAN port

    It definitely appears to be CPU limited though. I was logged in via ssh & monitoring top throughout the test and the CPU usage of "ksoftirqd_CPU0" was over 90% the entire test. That combined with the few other processes using a little CPU here and there pretty much had the thing pegged out. Interestingly, disabling the firewall seemed to make no difference.

    But I'd be surprised if the bandwidth of your internet connection is overwhelming the processing power of your router unless you have it doing a lot of other things.

    In addition to the test above I also watched the CPU usage of "ksoftirqd_CPU0" on my v1.0 WRT54GS while doing a throughput test at & at the
    SBC Yahoo! DSL Internet Throughput Test and in those cases the CPU usage of "ksoftirqd_CPU0" never rose above 6%. That's with a throughput of 5Mbps down & 520Kbps up which is the limit of my DSL connection.
  3. Trollkarlen

    Trollkarlen Network Guru Member

    Yep, I agree. I didn't have anything else installed except the firmware and I always left it at the default settings, more or less.

    I haven't logged on and looked at the processes while performing the tests. Is there a telnet or ssh deamon installed in the standard Linksys firmware as well?

    If the performance drop isn't caused by CPU limitations, what else could it be? Larger buffers on the D-Link router? I'm afraid I don't know all the details about TCP/IP, but what if the packets arrive in the wrong order for some reason? Would it be possible for the router to put them in a buffer, re-order them when the missing packets have arrived and then pass them on to the LAN?
  4. 4Access

    4Access Network Guru Member

    Unfortunately no.

    The only thing I can think of is did you enable QoS on the Linksys? Maybe a improperly configured MTU? Beyond that I can't think of much. It might be interesting to see if there's any change by connecting a 100Mbps switch or hub between your modem and the WAN port of the WRT. (You could use the LAN ports on the D-Link.)

    The router should just forward the packets as it's usually up to the hosts to resequence any packets that arrive out of order. Besides, the WRT should have plenty of memory, especially the GS models.

    If you decide to install a custom firmware such as DD-WRT that includes ssh & telnet you can check the memory status with the "free" command.

    Well I'm kind of out of ideas. I'm assuming that when you do the tests you haven't recently been running P2P or other traffic intensive apps?
  5. Trollkarlen

    Trollkarlen Network Guru Member

    Thanks, I'll do some more testing tonight.

    I don't have QoS enabled but I did some testing with MTU, but in the end I decided to leave it at Auto as I couldn't see any improvements when changing it.
  6. Trollkarlen

    Trollkarlen Network Guru Member

    This is interesting: using the D-Link as a hub gives me almost the same performance as when it's used as a gateway. With the D-Link as hub I get 7-7.5 Mbps down but this drops to 5-5.5 Mbps if the D-Link is removed.

    I was only using the DD-WRT firmware this time. There wasn't any noticable difference in CPU utilisation with or without the D-Link as a hub (the most CPU intense process was "ksoftirqd_CPU0" but it never used more than 30%). If I change my config back to use the D-Link as a gateway "ksoftirqd_CPU0" disappears completely from the top list and I get 8+ Mbps down.

    Memory utilisation seems low as well with approx. 20 MB free.
  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice