Wan Mtu Vs Lan Mtu

[Bob Carrick]

Nope, but I'll bug them this week.

[BlackSword]

^Bump

[Kirby Smith]

In another thread, SG TCP Optimizer Beta 6a, rmrucker wrote:

++++++++
WinPoet is equally quirky. It uses a "TunnelMode" value which whacks several bytes off the packet size for a "Tunnel Wrapper" in addition to the PPPoE wrapper. On some ISP's, the wrapper is essential to the connection and cannot be removed. BUT, more often than not, this Tunnel Wrapper is not needed and can be removed -- recapturing those lost bytes.
++++++++

I am on Verizon DSL and was initially supplied with WinPoet. As I am now operating successfully with a Nexland router in place of WinPoet, obviously a _particular_ Tunnel Wrapper is not being generated by WinPoet, and presumably therefor not needed by Verizon. However, if Verizon expected those bytes to be there wouldn't that limit the Verizon PPPoE user to an MTU less than 1492, even if the Nexland router could be configured to operate with an MTU higher than its predefined 1472 ???? Not that I see how Verizon could strip out bytes that weren't packed there first by the router. Maybe Verizon can find the PPPoE header whether or not the Tunnel Wrapper is present. Otherwise, I don't see how Verizon can work with some customers using WinPoet, and others not. ???? Sorry I don't have the means at hand to force various MTU values on Verizon to see what MTU value above 1472 causes the interface to gag.

kirby

[Bob Carrick]

If your MTU is too big it will still work, it will fragment though, which can cause poor transmissions, but it will still work and you may never even notice.

[Kirby Smith]

Bob:

My question doesn't relate to fragmentation, but to how Verizon's servers manage to deal with packets which may or may not have this Tunnel Wrapper included. Does the packet just get passed along to the ultimate destination, say, Speedguide.net, where the wrapper is somehow ignored?

kirby

[Bob Carrick]

I can't comment, because I've never heard of a tunnel wrapper in the PPPoE specification.

[Kirby Smith]

I don't think the absence of a "tunnel wrapper" in the RFC necessarily keeps DSL providers from inveigling Wind River into adding such a "feature." I imagine you can question "rmrucker" more concisely on this subject than I can.

https://www.speedguide.net/forums/ ... genumber=3

kirby

[Bob Carrick]

I will be getting an answer from the head at FinePoint who have the licensing rights to built and distribute WinPoet now.

[rmrucker]

A couple of things to think about:

1) Many routers -- especially Linksys -- used to force the MTU to 1476 on PPPoE connections. After many futile attempts to get the answer as to why this was the case, Linksys finally modified their firmware and bingo, the MTU was 1492. Many other router companies (e.g., U-Gate) seemed to do the same thing -- were they simply following Linksys? I do not know.

The modification is NOT for all packets, but only for TCP-IP. Therefore, a user could "ping" a normal size ICMP packet (1492), however, a TCP-IP packet was limited to 1476. To the best of my poor memory, the modification that was made by the router was something like this:

If a packet with a SYN flag set had an MSS value that was too high, the router over wrote the MSS to 1436 -- therefore limiting the MTU to 1476.

This is apparently controlled by the firmware as it was completely dependent upon what firmware the router was using.

2) EnterNet / Access Manager limits the MTU to 1454 -- but not because it uses a "wrapper". Instead, it uses a mis-named registry entry ("MaxFrameSize") to limit the packet size to 1454. If you modify this setting to 1492 (slightly more complicated than this), you can use EnterNet just like any one else. Therefore, increasing the MTU size does not prevent you from using EnterNet -- the smaller packet size is not a requirement.

The response from EnterNet was something like "1454 is the best packet size for PPPoE". Their explanation was never satisfactory to me...

3) WinPoET has a "TunnelMode" registry value. Open regedit and search for that term. If this is set to "1" (yes), then as best we can tell a 28-byte "wrapper" is applied to the packet -- limiting the MTU to 1472. This certainly may be an 8 byte PPPoE wrapper and then an additional 20 byte "TunnelWrapper". As I have never actually used PPPoE, all of my information is second-hand.

If you change this entry to "0" (no), then you can use a 1492 packet size. The problem is that some people have tried this and found out that they no longer can connect to the Internet. So some connections require "TunnelMode" and others do not. I could never get a decent response from WinRiver as to what exactly the TunnelMode was used for. I had a few guesses, but no solid answers.

[Kirby Smith]

Thanks rm. Fascinating. And my education via this thread just keeps going and going and.....

kirby

[rmrucker]

I modified my post to update the numbers to be correct after I reviewed some of my old records. The Linksys MTU phenomenon was quite weird and was a little more complicated than I stated above. Different firmware versions caused slightly different responses, but the most bizarre problems were in the 1.36-1.37 versions. They used this strange scheme to modify the MSS field in packets with the SYN bit set:

If MSS < 1437
- no change

If MSS 1437-1452
- over write MSS to 1436

If MSS > 1452
- over write MSS to 1322.

As above, the more recent firmware versions do not have this afflication.
______________________

I looked through the rest of this post and I have some more comments for you to consider. MTU problems are the most challenging of all!

What can screw up your MTU?

1) Inside your computer
..... a) NIC MaxMTU setting is incorrect
..... b) PPPoE Software quirk (EnterNet, WInPoET, etc.)
..... c) Internet Sharing Software (MS-ICS, WinGate, etc.)
..... d) Other -- VPN, IPSec, etc.

2) Outside your computer
..... a) Router/hub
..... b) Your gateway
..... c) Another router in the path with a low MTU

Then you can break down low MTU problems if they are:
1) Limited to TCP-IP
2) Affecting all packets
______________________________

So, what I would like to know is this:

1) If you remove the router and use a software PPPoE, what is your MTU for TCP-IP?
2) With and without the router, what is the largest payload you can ping with?

[Kirby Smith]

I was so impressed by the communication stability improvement with the Nexland router over WinPoet I hoped to never use WinPoet again. Fortunately for the future of this proposed test, WinPoet is still installed on my home machine, just disabled, although I will have to rediscover just what I switched off. At the moment I recall three parts, a service part I disabled in the service menu, a desktop dialer part, and a startup executable part I disabled using startup.cpl. I don't know if the dialer part is needed for pinging.

The bad news is that I may not be able to do this experiment for some time, as this week and weekend are pretty much fully committed. Also, I do not know what MaxMTU, etc., keys WinPoet may have previously set up before I started using the router, nor which related keys were absent before I started messing with the Optimizer. I am unclear what of the registry is relevant to WinPoet and whether any test I ran now would be equivalent to that with a clean install; perhaps WinPoet ignores the key values and does its own thing.

I am also not really familiar with changing ping packet sizes, although I vaguely recall there being something like that in the Optimizer. I will have to check. Also, I should look at the help response for the ping command, it may allow setting packet size. Perhaps the Optimizer will override whatever WinPoet thinks it is supposed to be doing. Edit> I see on my NT4 machine at work that the console ping command allows setting of the key parameters. And, to my surprise, 1472 is the max unfragmented size here. I doubt that between me and speedguide.net there is a WinPoet connection, but there are a pile of switches, routers, servers, frame relays, T1 lines, etc. <Edit

As you can tell, I have a lot of implied questions in the above rambling. Feel free to comment.

kirby

[Kirby Smith]

I have another instance of Win2k on the computer that is available for an emergency repair by changing the boot order. I don't think this instance has had any Ethernet meddling. I can install WinPoet there and test while leaving my main Win2k environment undisturbed. This will require also installing my firewall, etc. I see a good part of a day hacking here.

The NIC MaxMTU is another issue. Your mentioning it suggests that packets are formed in the OS, then reformed in the NIC, or in the NIC driver software. I haven't deliberatly modified any NIC parameters, unless modification happens as a result of using the Optimizer. I'm not even clear where they are modified under Win2k. I thought the driver took its parameters from the keys the Optimizer messes with. (My recollection is that my NIC is a D-Link 530TX+.)

kirby

[rmrucker]

You don't have to bother doing this unless you want. If you look at the big picture, what is the diffence between a 1492 byte packet and a 1472 byte packet? It's about a 1-1½ % difference. How much difference is it really going to make in your speed? Probably not tons.

So after you go through the motions and finally get your MTU to 1492, don't expect things to change drastically. I would be happy to figure it solely because it is fun and challenging. You may not be so inclined.

Assuning you are interested, you may wish to install RASPPPoE and use that instead -- if you have any problems with getting WinPoET back up. Don't worry about the registry or the Optimizer -- I can walk you through all of that.

The Optimizer does allow you to "ping" fairly easy. You can also use a DOS window and ping using the instructions here: http://www.speedguide.net/Cable_modems/ ... nced.shtml

[Kirby Smith]

I had actually tried the ping commands I hadn't previously known about on my work NT4 PC last night. I thought I had sent an addendum to this thread on the results. In a system that has nationwide LANs, WANs, routers, frame relays, T1 lines, etc., but no PPPoE links I assume, the max ping packet size is 1472. Mind boggling.

I agree fully that 1472 is fine. That is what I am running on my home computer that started this thread. I would never bother with the process we are describing for a mere 20 bytes per packet. The whole point of this exercise is to determine what the "truth" is about PPPoE packets. In particular, why some MTUs have to be, or are for obsolete reasons perhaps, less than 1492. My anticipated (but not too soon) testing along the lines you suggested is intended only to determine if Verizon DSL will work above 1472. But my main interest is in the information that Bob is weedling out of the key players.

kirby

[Bob Carrick]

Yes, I'm now also waiting to find out about WinPoet and it's MTU and why etc...