Kitz Forum

Broadband Related => ADSL Issues => Topic started by: PhilT on April 10, 2006, 07:23:07 AM

Title: BT MTU
Post by: PhilT on April 10, 2006, 07:23:07 AM
On the tweaks page you say that "Since then, BTw have carried out work on exchanges to set the adsl network with the MTU values of 1458."

I think this is misleading as the work to fix the original problem has left the BT network able to cope with MTU values of 1500 (and higher) in other words the "need to be 1458 due to BT" is now ancient history and ought not to be the first thing one sees on the tweaks page.

Some ISPs may have MTU restrictions, but others (Demon for example) and BT Wholesale do not.

Cheers,

Phil
Title: BT MTU
Post by: kitz on April 16, 2006, 02:47:56 AM
Hiya phil
Good to see you on here :)

Thanks for bringing that to my attention, I wrote that about 2.5 years ago, and you are quite right it certainly does need updating. :/
In fact the MTU pages is due an overhaul anyhow, so I will look at doing that asap.

Sorry for the delay in reply as Ive been away for a few days.. and I'll update the MTU page as soon as Ive caught up on other the other stuff thats been going on over the past week.
Title: BT MTU
Post by: feliscatusx2 on August 07, 2006, 08:04:03 PM
Any news on this yet?

I'm using MTU of 1478 & RWIN of 51768 on a non Max 2Mb connection.  PCPitStop recommends a RWIN of over 60K, but then they are Americans

Tune Up utilities claims MTU of 1492 & RWIN of 128000 is best.  

TCPOtimizer suggests MTU of 1500 & RWIN of 128480.

Should I worry, should I care?

Does it really make that much difference?????

Any illumination gratefully recieved
Title: BT MTU
Post by: kitz on August 07, 2006, 10:19:12 PM
The page that phil referred to got a complete overhaul a couple of days after I made that post... and was renamed to The history of tweaking (http://www.kitz.co.uk/adsl/tweak.htm). The bit at the bottom "tweaking today" expresses my own person opinion.

I also did a brand new page MTU and RWIN explained (http://www.kitz.co.uk/adsl/MTU2.htm)

- Gosh just noticed that I obviously wrote that before I was maxed (which was about the 20 somethingth of April).
My opinion still stands..  I still tweaked when maxed, and it did make a difference on my connection. I was shocked to see an almost 400kbps increase by changing my rWIN.

I spent a good few hours in the early hours of the morning, testing and playing with a few settings, and on a couple of machines, before I was happy with what was best for my connection.

Best results for my maxed connection are at the bottom of the 8Mb table
http://www.kitz.co.uk/adsl/tweak2.htm

Funny enough I noticed the 2 days ago that the machine I was using was untweaked, so I changed it.. and got a slight boost on speed on there too.

Must admit though now.. since more+more people have got max and speeds arent as stable as they first were in the early weeks.
In the wee small hours I have had quite a few 7.09 or 7.19Mbps results which is I think is just about the theoretical maximum.

My best ever was this one..  but it was a one off and perhaps a fluke cause its a bit more than you are supposed to be able to get.  The speed test I did immediately after was 7.19Mbps.


(https://forum.kitz.co.uk/proxy.php?request=http%3A%2F%2Fwww.kitz.co.uk%2Ftemp%2Fbestever.jpg&hash=ddc0a6634cef3ecf51bf54030fddcb10)

Theres some settings to try for 2Mb connections on my page.. Tweak settings (http://www.kitz.co.uk/adsl/tweak2.htm)
it all depends if you have time to play and find what is best for you.

As it stands now, I'm quite happy to say that 1430 MTU works for me, and the larger RWINS seem to work.  It now only takes me 2 mins to change the settings - so I do :)
Title: BT MTU
Post by: roseway on August 08, 2006, 08:10:47 AM
Strangely, these tweaks seem to make little or no difference to download speeds on a Linux system, although reducing the MTU value below its default 1500 does make a small improvement to upload speeds.

Actually, it's difficult to see why reducing the MTU below 1500 would make any improvement to download speeds. When downloading you're at the end of the chain, and so long as your MTU is at least as large as the biggest packet you won't be causing any fragmentation. I suspect that the fact that it does have an effect is a quirk of Windows.

Where RWin is concerned, by default Linux uses a dynamically variable value for both the Receive and Send windows. In theory at least it always optimises the values for best speed. You can override the defaults with fixed values, but it has little or no effect.

Eric
Title: BT MTU
Post by: soms on August 15, 2006, 11:59:39 PM
A very interesting result of the Windows MTU tweaks from Kitz's sites is a physical result - my router is much cooler!

It was mentioned that the new settings could reduce the load on the router - and it must have.
Before hand the router was getting very hot indeed, even when pretty idle and was always too hot to grip in your hand. The same goes for the power adaptor, before hand again in got worryingly hot.
Now both items are much cooler and I can sleep safe at night without fear of the router or adaptor starting a fire!

Also the number of connection problems has been reduced. I don't think the router was dropping out anyway, as it always stayed in sync but wouldn't do internet for clients unless disconnected/reconnected from the net using any method.
(I am guessing those probs must have been something else like virtual circuit channels, connection probs, overload or something.)

And of course in Windows there has been a noticable performance improvement for general internet usage.  :)

A truly useful article by Kitz which not only helps things run smoother but makes your equipment last longer at the same time  :o
Title: Re: BT MTU
Post by: hake on September 09, 2006, 07:36:47 PM

MTU size affects packet fragmentation.  I chose 1458 initially because of the BT limitations a while ago.  Since then, I have stuck with 1458 as a compromise setting with consideration to fragmentation.  A MTU of 1458 gives  a MSS of 1418.

I experimented with ping -l MSS -f

I determined that a MSS of 1418 and therefore MTU of 1458 (MSS+40) seemed to reduce fragmentation to negligible levels and I have used it eversince.

 ???
Title: Re: BT MTU
Post by: kitz on September 09, 2006, 11:09:56 PM
Yes this is one of those  "does it or doesnt it" debates.

It is harder now we have max to actually test the impact that tweaking has on our connections, more down to the fact that speeds tend to be more variable when on max.

I know for sure though that when on a 2Mb product that I did get slightly better speeds on a tweaked connection.
The testing I did when on max, was mostly playing with the rWin sizes..  and it was when I first got max and early one morning when speeds where pretty stable.
Now that more people have max..  I find it extremely hard to get the same speeds from one minute to the next unless it is the wee small hours.

I guess I just tweak now, because I can and it only takes me a couple of mins.
are there are still people who benefit from tweaking MTU when having problems with certain websites.
Title: Re: BT MTU
Post by: hake on September 10, 2006, 09:25:50 AM

Kitz, I upped my RWIN to 56720 (from 34032) and the difference was considerable.  Thanks for the advice (I cannot remember where I read it on tour site).

I understand that if RWIN is >= 65535, that some jiggerypokery is necessary with other settings (Window Scaling) that may not be beneficial.  I will look this up my TCP/IP tomes.
Title: Re: BT MTU
Post by: mr_chris on September 10, 2006, 09:40:45 AM
That's all that's needed - Window Scaling needs to be set to "on" to allow RWIN of > 65535

The reason it may not be beneficial is simply if there's even a small amount of corrupt packets resulting in retransmitted data, the larger the RWIN, the more data that needs to be retransmitted.

Therefore if you find yourself downloading something from somewhere that seems to be losing a lot of packets, your speeds may well be better with a smaller RWIN because you're not wasting as much data.
Title: Re: BT MTU
Post by: hake on September 10, 2006, 10:13:55 AM

RFC1323 options need to be enabled for RWIN > 65535.  Windows 9x might also need the MS Vtcp386 fix.

I have reservations about very large RWINs which is recovery from packet loss.  A smaller MTU value reduces fragmentation and therefore the incidence of packet loss.  Van Jacobson seems to provide immunity from the necessity for a pipe refill in the event of single packet loss only.  Multiple packet loss results in having to refill the pipe so I would consider MTU size carefully when using large RWIN values.

I would reduce MTU to 1430 if using RWIN > 65535 as the profit in using large RWIN values is greater than the slight loss of efficiency in using a smaller MTU.  Nothing is gained from having to frequently refill the pipe.

There is no such thing as a free lunch (although I did have a free breakfast once).
Title: Re: BT MTU
Post by: roseway on September 10, 2006, 11:05:40 AM
Sorry if I seem to be argumentative, but a smaller MTU value won't reduce fragmentation of downloads, it will tend to increase it. If you have an MTU of 1430 and a 1450-byte packet arrives (say) then it's going to get fragmented when you receive it. With a higher MTU value this wouldn't be the case.

This is why I have so much difficulty understanding why MTU tweaking can speed up downloads.

Eric
Title: Re: BT MTU
Post by: mr_chris on September 10, 2006, 11:22:50 AM
The way I understand it is this:

MTU values are agreed by both ends at the beginning of a connection. By default, the Windows MTU is 1500. Most web servers can also deal with this, so a 1500 byte MTU is agreed.

However, because a router along the way may only be configured with an MTU of say 1458, each 1500-byte packet that comes down the line is split into two packets - one 1458 bytes and the other 42 bytes long. Obviously that incurs twice the header overhead for each 1500 bytes, thus reducing efficiency.

Path MTU Discovery is meant to work this out automatically, by finding out the maximum MTU that is sendable through the entire network. However, if along the way somewhere, ICMP traffic is blocked, the MTU reverts to simply whatever is agreed by each end, as above.

This is one reason why it makes sense to allow yourself to be pingable, because blocking ping on a lot of routers blocks ALL ICMP traffic, rendering PMTU discovery, and a lot of other invisible TCP/IP functions, useless.

Therefore, it's easier to lower your MTU to something like 1430 to avoid all these problems. It is a commonly accepted fix for when certain areas of Microsoft and Ebay's websites don't work, too (presumably due to issues with PMTU - neither of their websites are pingable!)
Title: Re: BT MTU
Post by: roseway on September 10, 2006, 12:50:42 PM
Yes, that makes sense Chris, thanks. My one remaining still-not-understood bit is that MTU tweaking has absolutely no effect on download speeds in Linux systems, unless you make the value really low (which reduces the speed of course). I guess that the Linux implementation works in a different way (or perhaps Path MTU Discovery works better).

Eric
Title: Re: BT MTU
Post by: mr_chris on September 10, 2006, 01:01:48 PM
> MTU tweaking has absolutely no effect on download speeds in Linux systems

Not come across that before, I'm not sure, to be honest. I can't see how it would be different from Windows to Linux.
Title: Re: BT MTU
Post by: hake on September 10, 2006, 01:29:55 PM

I only receive IP traffic from connections initiated by my own hosts.  MTU should consequently be what my hosts have negotiated with the hosts that my hosts are connected to.  The MTU should not therefore exceed the size negotiated.  All traffic from hosts I have not sent SYN packets to is rejected so the MTU of unsolicited packets is irrelevant.

That is my understanding of the MTU situation.

If I am wrong (and I often am), I always welcome advice correcting any errors in my understanding since I would rather stand corrected than delude myself that I am right.  TCP/IP is an intensely complicated subject so misunderstandings are inevitable and I rely on others to advise me when they believe that I am in error.

 :-[
Title: Re: BT MTU
Post by: roseway on September 10, 2006, 04:29:14 PM
If I am wrong (and I often am), I always welcome advice correcting any errors in my understanding since I would rather stand corrected than delude myself that I am right.
Yes indeed. It can be uncomfortable to have ones own cherished beliefs shot down in flames, but it's certainly better to find out the truth.
I've been known to be wrong too. :'(

Eric
Title: Re: BT MTU
Post by: mr_chris on September 10, 2006, 07:12:25 PM
As have I, many times!

Quote
I only receive IP traffic from connections initiated by my own hosts.  MTU should consequently be what my hosts have negotiated with the hosts that my hosts are connected to.  The MTU should not therefore exceed the size negotiated.  All traffic from hosts I have not sent SYN packets to is rejected so the MTU of unsolicited packets is irrelevant.

That's right... it will never exceed (a) what you have set on your PC, and (b) what the other end negotiates with you during startup handshake. However there could still be a router somewhere along the line with a lower MTU value which is also blocking ICMP, causing Path MTU to fail. This would then cause packet fragmentation, or worse still, no traversal at all (which is what seems to happen with ebay and microsoft). Lowering the MTU helps overcome this.

You seem to have a pretty sound understanding of TCP/IP and how it works. As you say, it's an insanely complex subject and there's far more to it than meets the eye!
Title: Re: BT MTU
Post by: hake on September 10, 2006, 11:48:41 PM
I like this forum.  I expect to be lowering the tone of it for some time to come.

I like to know how things work.  TCP/IP is a very interesting subject because it is very likely to be the basis of future worldwide communications and because it is very intellectually demanding, especially because it is so dynamic a system.  I will be giving RFC1323 a serious bashing this week.  I might even understand some of it!!!

I won't be making my RWIN more than 65228 (=1418*46) because I have found that the 5+MB bandwidth that adsl max has just given me seems to be more than the capacity of most servers to fill, except for Microsoft's: I downloaded WinXP SP2 (266MB) for fun and the MS server belted it out at 4.5MB.  Most servers seem to be limited to 2MB or less (often considerably less) so I am happy with RWIN appropriate to 2MB.  The fast connection enables my wife and me to surf without slowing each other's traffic.

I have already found the contributions made by forum members to be very valuable to my efforts to further my knowledge and I hope to put something back.

Kitz's RWIN tips really added pzazzz to my adsl connection.  I like it when things work just perfectly and I have a lovely Internet connection now.

I'll shut up now since my posts are becoming irrelevant to this thread.

 :doh:
Title: Re: BT MTU
Post by: kitz on September 11, 2006, 09:14:22 AM
>> I like this forum.  I expect to be lowering the tone of it for some time to come

and welcome..  we're quite a friendly bunch when you get to know us.


psssst just watch out for those mods roseway and astral in the chat section.. 
I just leave em to it sometimes and go... huh??  :-X
 

>>> I won't be making my RWIN more than 65228
You could go on for ever with tweaking trying different settings..  Im now at the point when.. ok this seems to work well for me.. and I leave it at that.

>>I'll shut up now since my posts are becoming irrelevant to this thread.
hmmmm  you should fit in quite well on this forum then.. 
- see above comment- we're already quite good at doing that :lol:



Title: Re: BT MTU
Post by: roseway on September 11, 2006, 10:14:59 AM
Quote
psssst just watch out for those mods roseway and astral in the chat section..
I just leave em to it sometimes and go... huh??

I have no idea what you mean. 8)

Eric
Title: Re: BT MTU
Post by: kitz on September 11, 2006, 10:23:06 AM
huh??


 :-X
Title: Re: BT MTU
Post by: roseway on September 11, 2006, 12:55:58 PM
I was only joking, Kitz. :) :) :) :) :) :) :)