SuperCharged EVDO - Free Performance Tip

Three60guy · EVDO User Joined: 21 Feb 2007 Posts: 73

When I use 1465 it says it needs to be fragmented but when I use 1464 it doesn't. So I should use 1492? Presently it is set at auto with 1500 grayed out.

Fox McCloud · EVDO Junkie Joined: 02 Dec 2006 Posts: 563

mattengland · Posted: Thu Jun 28, 2007 8:12 am Post subject:

I have a variety of questions stemming from my confusion on this topic:

SG TCP Optimizer tells me, when I visit the "Largest MTU" tab that I should set my MTU to 1500. Then I run the ping/analyzer, and it recommends a MTU of 1492. My manual ping testing suggests a number of "1464," but I don't know what thing this "1464" represents. Is it the MSS (it's not clear to me after reading the above posts)? And further, does 'MTU = MSS +40' or is it 'MTU = MSS + 28', for I believe I'm reading both suggestions. The SG TCP Optimizer seems to like adding 28 (judging by 1492 - 1464 = 28 as per my laptop).

Additionally, after setting my MTU to 1492 (as per SG TCP Optimizer) and rebooting, the SG Analyzer tells me my MTU is 1452.

Also, the SG TCP Optimizer seems to suggest that an MTU setting is per-netowrk-adapter; I checked the box for "all network adapters," but the EVDO adapter is not in the "Network Adapter selection" pulldown list. This does not give me confidence that I'm truly affecting my EVDO's parameters, nor that the setting is actually "taking" correctly (given that DrTCP and TCP Optimizer tells me my MTU is 1492 and the SG web-page analyzer tells me the MTU is 1452).

This leads me to suspect that we, or at least I, are comparing apples to oranges here. It could be that there are actually multiple different definitions and settings for this thing we are calling "MTU" and/or there could be a different MTU setting in different contexts (eg, one for my wireless ethernet, one for cabled ethernet, one for EVDO, etc). In any case, there's something that's not adding up. Thoughts?

To reiterate, I'm still left to understand the proper relationsthip of MSS to MTU (is it a difference of 28 or 40 or somethign else)?

Next topic:

MTU-to-RWIN relationship. How does one evaluate a properly multiple (assuming RWIN is best a multiple of MTU)? Is it trial and error? Is there a general, qualitative tradeoff for increasing or decreasing the RWIN? eg, does one gain bandwidth performance while possibly degrading latency performance by increasing RWIN? I have yet to see anyone speak to this.

Also, why to MSS's vary so much from differnet Windows systems?

All in all, one might possibly understand how this issue could be a bit cumbersome for those new to the topic, even for those who are technically savvy.

By the way, the ping command above is not formatted properly (it has an extra space that breaks the syntax and could confuse some novices), it should be this:

mattengland · Posted: Thu Jun 28, 2007 8:16 am Post subject:

mattengland · Posted: Thu Jun 28, 2007 8:22 am Post subject:

Another question: why does software not exist to support a "one-click" optimization of all these settings so that people like me don't have to do all this work an analysis?

Maybe a one-click setting is to general? Maybe a "click this button for best bandwidth" and "click this button for best latency" would be most useful?

All in all, I'm rather stunned that this needs to be this hard for a user and that someone has yet to put something together to automate all this. Further, this issue seems to have broader implications than just EVDO.

clarkgable · EVDO User Joined: 14 Sep 2005 Posts: 89

mattengland · Posted: Thu Jun 28, 2007 9:07 am Post subject:

I feel I should summarize my questions about the procedure for optimizing Windows network performance (which is what I assume we are doing here):

1) Determine the optimized MTU for your system (assuming there is one and only one "best" MTU for an given Windows instance? This gets to the tradeoff question), and set the value with an appropariate tool.

2) Based upon this "best" MTU, set your RWIN (aka, "TCP receive window size") as some multiple of the MTU.

Are these the 2 basic steps that one should be performing in this network-optimization procedure? Or is there more (or less)?

As to what this MTU-to-RWIN multiple should be is also a point of confusion that I'm not sure I'm clarified earlier.

For example, here's a quote from my SG TCP Analyzer web page:

mattengland · Posted: Thu Jun 28, 2007 10:38 am Post subject:

I've created:

http://en.wikipedia.org/wiki/Windows_single-user_networking_performance_optimizing

I'd like to see us start driving resolutions/summaries/conclusions/references there as a community-editable place to get all the answers, at least for:

http://en.wikipedia.org/wiki/Windows_single-user_networking_performance_optimizing#EVDO

Also note that in the above wiki article that I've discovered the dslreports.com Tweak tools and faqs (see the article).

Please keep the discussions here and not on the wiki system (I've pointed the article to this discussion thread in the references section).

Also, here's a quote from a friend of mine with networking-industry experience:

mattengland · Posted: Thu Jun 28, 2007 7:22 pm Post subject:

While I wait for responses to my above posts--even if it's someone saying that I have no clue, it would be nice to get a response--I did a tad more testing, with little if any further clarity.

I took a stab at trialing some of the windows-networking optimizers listed at download.com's Dial-up & Connectivity section and had no epiphany. However, none of the tools I've seen thus far seems to be "EVDO aware," or at least list a networking type of "EVDO." That concerns me.

I also have seen evidence that the MTU/MSS settings seem to be independently set for each network interface (eg, wireless ethernet, cabled ethernet, evdo seem to be separate interfaces with different settings/controls).

I'm still not sure which value, MSS or MTU, the "XXX" in 'ping -f -l XXX www.google.com' represents (and I'm looking for clarity on this from the community here), but this link suggests it's the MTU:

http://support.microsoft.com/kb/319661

If so, then at least one automated tool, SG TCP/IP Analyzer doesn't seem to calculate the same thing; SG TCP/IP Analyzer always says 1500/1460 for MTU/MSS for my evdo connection (but something else when I'm running wireless ethernet). However, when I run 'ping -f -l XXX www.google.com', XXX for me with the evdo connection is 996 (and if I make it bigger I get ping timeouts, not the fracture error message), and it's different on wireless-ethernet tests. So why is SG Analyzer telling me to set MTU to 1500? Further, I see no other optimizing utility setting the number to 996 either (although I'm not sure any of the other optimizers actually tell me what they attempt to set). Another mystery/mysteries unexplained.

Further SG TCP/IP Analyzer has this as-of-yet-unexplained MSS-to-RWIN relationship suggestions, eg:

mattengland · Posted: Thu Jun 28, 2007 7:32 pm Post subject:

The fact that my 'ping -f -l XXX www.google.com' commands hang with larger XXX's seems to be a problem. Here's a run from SG's optimizer:

mattengland · Posted: Thu Jun 28, 2007 8:04 pm Post subject:

Ok, so http://www.dslreports.com/faq/tweaks was interesting, particularly:

http://www.dslreports.com/faq/695

but I still don't see anything which helps me to resolve my fundamental issues (although it is another data point that the TCP header is 40, and the ICMP header is 28, which may resolve the 28/40 discrepency if so...although this faq is also not consistent on this point, I'm just reading from the "magic number" table).

In particular, I don't know why ping is timing out above XXX = 996. I see this text in http://www.dslreports.com/faq/695 :

mattengland · Posted: Thu Jun 28, 2007 8:31 pm Post subject:

fyi, I made a Usenet/newsgroup post at microsoft.public.windows.networking.wireless (and I copied another newgroup with the same post) overviewing this issue, asking for help. One might check there to see if anyone has input.