Question about Mac/PC internet speed

Discussion in 'Hardware' started by bumblebuzzard, Dec 9, 2007.

  1. What about the RWIN of XP? Since Mac's aren't subjected to the Receive Window hangups, perhaps a good ole Dslreports.com Tweak Test then running Dr TCP and resetting the RWIN and rebooting would do the trick?
     
    #21     Dec 10, 2007
  2. Well, that's a good suggestion, except that I already had the RWIN set to 513,920 and the MTU set at 1500, so I wasn't sure that a bigger window would make much difference.

    But since I wasn't sure it wouldn't either, I went to dslreports and ran their Tweak tool, according to which I should set the MTU at 1500 and the RWIN between 94900 and 251120.

    (I mean that's what it told me after I figured out that I was running on the DSL instead of the Cable and got it all straightened out and rebooted about 8 times and had a bourbon and branch, which is probably what I should've done in the first place instead of whatever it was I did do.)

    So I set the MTU at 1453, because that was the largest unfragmented packet size and was almost 1500, and the RWIN at 247,010, because that was 170 times 1453, which was sorta dumb because the RWIN is supposed to be some multiple of the MSS not the MTU, but, then I didn't know what the MSS size was, so I thought 'what the heck'. The result was that my speed went back down to about 4500 kbps.

    Then I reset it all to where I began, because those were the best numbers I had and I didn't feel like watching the thing reboot one more time. The speed went up to 4900 kbps.

    So, I guess it didn't work. Well, I didn't really ever increase the window size, but I'm pretty sure that's not the problem. Meanwhile, the G5 is still connecting at 7500, and I'm inclined to think that MuddBuddha may have the right answer along the lines of the Mac OS just handles UNIX better. I'm not actually sure what that means, but it sounds right to me.

    One thing all this fooling around does suggest is that connect rates can vary quite a lot based on the user's box, and it isn't always the ISPs fault. I would have expected some differences between the PC and the Mac, but the size of the difference is surprising. (To me. Not to my wife.)
     
    #22     Dec 10, 2007
  3. Well, you're definitely on the right track if you wanted to keep messing with it. But a RWIN that big is not better. It just ends up causing packet loss and resends. But that's why it's a crap shoot for your particular system.

    That said, I do believe there is an FAQ on dslreports that takes your ping rate to a site and then doing some math, comes up with an RWIN that's 'more correct' for you.

    But if you're done with screwing around with it, you're done :p
     
    #23     Dec 11, 2007
  4. Naw, I was just done for last night. I'll see if I can get a better parameter after the market closes today.

    Thanks for the help.
     
    #24     Dec 11, 2007
  5. xiaodre

    xiaodre

    Bumble, please may I ask you to do one more thing: please would you switch the ports on the back of the modem/router that the two machines use, so that the windows box is plugged into the port that the mac is usually plugged into and vice versa?

    This is the last of the basic troubleshooting steps (although you've already moved on to advanced troubleshooting of windows), and what this would do for me is it would rule out something going on with the modem/router, and point the finger directly at the windows box, which is where I suspect the problem lies anyway...
     
    #25     Dec 11, 2007
  6. xiaodre, I tried all four ports on the modem individually, and that didn't make any difference, so it does look like the windows machine is where the problem lies. Of course, I'm not sure it is a problem in the sense of something wrong. It may be working just as it should and just can't go any faster.

    I'm going to fiddle with the RWIN again as steve suggested. Maybe I can break something loose there.
     
    #26     Dec 11, 2007
  7. another question, are you using one of those cheapass onboard network card or a good one?

    good ones are generally 3coms, IntelPRO.


    if you have a el cheapo realtek or other inferior brands, here lies the issue.

    try newer updated drivers or beta drivers from the manufacturer website,
     
    #27     Dec 11, 2007
  8. Hmmm, yes, I suppose that could be a problem.

    I've got two LOM network controllers. One is part of the NVIDIA nForce4 SLIx16 chipset. The other is a Marvell Yukon 88E8053 PCI-e Gigabit Ethernet Controller. Both return the same results on the speed test.

    What do you think?
     
    #28     Dec 11, 2007
  9. Ok, for all of you on the edge of your seats wondering, "Will bumble really speed up his internet connection?" I've tried a number of different RWIN sizes and here's what happened.

    First, I should say that the "correct" window size seems to be an estimation. There are different suggestions as to how to calculate it. Dslreports has a formula which is (Avg latency * 1.50 * Available Bandwith)/8. Another calculation is just (Bandwith * Latency). Since average latency is a guess and bandwith is a guess and the formula is a guess, the calculation is a guess about a guess multiplied by a guess. The point being that this is not an exact science. Fortunately, as far as I can tell, RWIN size is not a very critical number.

    Dslreports' Tweak Test resulted in a recommendation to set MTU at 1500 and RWIN somewhere between 94,900 and 251,120 bytes. However, MTU is best set at the largest size before packets start fragmenting, and in my case that is 1453. Some people recommend that RWIN be an even multiple of MSS, which in my case is 1452. Running ping tests resulting in an average latency of 128ms. Pinging the server I'm using for all speed tests returned 25ms.

    With this hodgepodge of data, I started changing settings. My starting point was RWIN of 513,920 bytes and MTU of 1500, which are the numbers recommended by the Speedguide TCP Optimizer to which I was referred by silvermotion. The starting download speed with these settings was 4900 kbps.

    I tried setting RWIN down to 75,000 and up to 1,000,000 and many places in between. The result in every single case was a reduction in the tested speed to between 4400 and 4600 kbps.

    So my experience is that only two things have had a positive effect. First is the rebuilding of the TCP/IP stack, which I did with the XP utility; and second is the setting of RWIN as recommended by the TCP Optimizer. These two together resulted in an increase of about 400 kpbs, about 9%. This is worthwhile, but doesn't even put me in striking distance of the Mac.
     
    #29     Dec 11, 2007