When a 1500 MTU is not 1500 a conundrum

[LesD]

Not wishing to hijack another thread where MTU's of 1458, 1430 and Vista are mentioned the topic touched a nerve.

From way back when I was on fixed rate broadband I was advised to set my MTU to 1458.
When I moved to adaptive rate broadband I saw recommendations that my MTU should be set to 1500.

Namely here:  http://www.speedguide.net/analyzer.php?
Where I currently see this:

MTU = 1458
MTU is not fully optimized for broadband. Consider increasing your MTU to 1500 for better throughput. If you are using a router, it could be limiting your MTU regardless of Registry settings.

I have both before & since joining IDnet done the packet size ping test both at speedguide.net and manually from the Command Prompt and a MTU of 1500 looks right.

So I set 1500 initially in my Netgear DG834 (Mk1) router and subsequently in my new 2Wire 2700HGV router and ran TCPOptimizer (and others - DrTCP etc.) to set the TCP parameter to 1500. Did the Restart and after this type of reboot all looks good with the Speedguide Analyzer. (I get the nice "green" message for those that have seen it)

Switch the machine off at the mains for a short time and reboot from "cold"  and the SpeedGuide Analyzer comes back with a message like this again! 
MTU = 1458
MTU is not fully optimized for broadband. Consider increasing your MTU to 1500 for better throughput. If you are using a router, it could be limiting your MTU regardless of Registry settings.

For me this is predictable, repeatable, demonstrable and downright perplexing to the point where I have been right though the registry looking for 1458 in decimal and hex and put every instance I have found to 1500 decimal or hex as appropriate, all to no avail.

When I got the 2Wire I tried again with MTU = 1500 set in it and ran TCPOptimizer to get the TCP to be 1500.
Shortly after that I got some of the lowest evening speeds I had encountered since joining IDNet so I put everything back at my tried and tested MTU of 1458 and all seems well except Speedguide's Analyzer that thinks my MTU is wrong!

I have a theory that Speedguide think that we all live in the good old USA and have probably never heard of BT but I am good at inventing red-herrings like that! 

Can anyone shed any light on this little mystery for me?

__________________
Regards,

Les. Denning


P.S. My system details are:

Microsoft XP Pro +SP2
Intel Pentium 4 640 3.2GHz 800 FSB LGA 775 CPU EM64T +2 MB Cache
Gigabyte P4 Titan Series Motherboard GA-8S661FXM-775 Rev: 1.1
2 GB PC3200 DDR 400 CL3 Memory
AOpen NVIDA GeForce FX5600 256 MB DDR AGP 8x Graphics Card
2Wire 2700HGV Dual SSID ADSL/Firewall/Router

The current SpeedGuide summary for my settings is:

« SpeedGuide.net TCP Analyzer Results »
Tested on: 04.10.2008 15:14
IP address: 91.135.xx.xx

TCP options string: 0204058a0103030201010402
MSS: 1418
MTU: 1458
TCP Window: 260912 (multiple of MSS)
RWIN Scaling: 2
Unscaled RWIN : 65228
Reccomended RWINs: 65228, 130456, 260912, 521824
BDP limit (200ms): 10436kbps (1305KBytes/s)
BDP limit (500ms): 4175kbps (522KBytes/s)
MTU Discovery: ON
TTL: 51
Timestamps: OFF
SACKs: ON
IP ToS: 00000000 (0)

[Steve]

Cant shed any light but since moving to idnet that site has always quoted my MTU as 1460 no matter what higher value I tried to set it at. However if memory serves me right it quoted "correctly" when I was with ***ex. Now I am using vista I have not bothered changing these values.The router has always been set at 1500.

[LesD]

since moving to idnet that site has always quoted my MTU as 1460

I had 1458 from SpeedGuide when I was with Tiscali but you are right I recall that it was 1460 after I moved to IDNet but only with 1500 set in the 2Wire. Now I have 1458 everywhere the router setting will be the limiting factor I guess.

[Simon]

Sorry Les, can't really help, but my 2700HGV is set to 1500, and all seems to be working OK.  I did dabble with TCP Optimiser some time ago, and all it seemed to do was make things worse, so I ended up just reverting to the default settings.  With things like this, I tend to fall back on the "if it ain't broke..." mentality.

[Sebby]

I wouldn't take too much notice of that site, Les. Perhaps reboot your router to ensure that the MTU setting takes effect (ensure the value is still there, too). If you're 100% it's set to 1500 on XP as well, you're all set.

The best way to set MTU from that point is the ping test. Once you've got the final figure, I would set it on each PC (leave the router as 1500) using Dr TCP.

[Rik]

Hi Les

Usually this sort of thing occurs when you're routed via a circuit which has a smaller MTU in place, or when the auto-detect facility isn't working correctly. It's nothing to worry about, though. Different ISPs recommend different values of MTU (it really shouldn't change between fixed rate and Max, though). Orange used to use 1430, Nildram 1458, AOL something ridiculously low. On IDNet, 1500 or 1458 will both give excellent results, and you can always use TCP Optimizer to check what value will work best for you.

[LesD]

On IDNet, 1500 or 1458 will both give excellent results, and you can always use TCP Optimizer to check what value will work best for you.

Thanks Rik,

Having been round this bouy a few times I think I will stick with MTU = 1458.
I also believe that if you follow through the calculations from this maximun transmission unit to the size of the RWin buffer that as it is buffers that we get to it is best that size wise they match so that there is no tendency to over-fill or under-fill them with any ensuing complications that this may lead to. For this reason with the TCP MTU at 1458 I shall keep the Router at 1458 too. This is my preference and it may not be critical but is has the right feel to it for me.

In the summary that SpeedGuide provides it says that I have, "MTU Discovery: On".
My thoughts on this were that maybe this is why when I set the TCP MTU to 1500 a bit later SpeedGuide detects it as something else. That is possibly after the "Discovery" function has done its bit but I am out of my depths now despite having been "round the bouy few times"!   

[Rik]

MTU Discovery means that the system should adjust as needed. In my experience, this seems to result in the site reporting the smallest MTU discovered along the way, rather than your MTU. Do what you're doing, stick with what feels right to you - the adjustments generally only produce small changes anyway.

[MoHux]

You may find that reducing your router MTU from 1500 will also reduce your upload speed!!

Mo

[LesD]

You may find that reducing your router MTU from 1500 will also reduce your upload speed!!

Thanks for that pointer Mo I have noticed some consistently lower upload speeds lately eg:



But my Netgear was pretty much always set at 1458 and I saw upload speeds of well over 300 kbps with it set at 1458.

More food for thought!

[Rik]

That upload does look low, Les. What do you get at:

http://www.speedtest.bbmax.co.uk/

[LesD]

This is the BBMax result from a moment or so ago but I always think their results can be a touch on the optimistic side and the time stamp is always iffy:

Results for speedtest    #1207938501
Date of Speed Test: 2008-04-11 19:28:21
Download Speed: 3400 kbps (425 KB/sec transfer rate)

Upload Speed: 356 kbps (44.5 KB/sec transfer rate)

I would do more thinkbroadband ones if it wasn't for the circa 8 MB a shot!

[Rik]

I suspect the BBMax upload is closer to the truth than Speedtest.net, Les. I find it tends to agree pretty closely with the TB results.

[Sebby]

I agree; I've always found speedtest.net to be quite inaccurate.

[LesD]

I suspect the BBMax upload is closer to the truth than Speedtest.net

On reflection I must agree because currently connecting at 4352 kbps I have an IP Profile of 3500 kbps so Speedtest.net's 3540 kbps downstream has to be suspect!

OK guys see Sebby agres so will I have to admit that I only went back to Speedtest.net when I saw a lot of folks on this forum publishing its result. I jumped to the conclusion that it sort of had this forum's approval but I guess I was wrong!

[MoHux]

Thanks for that pointer Mo I have noticed some consistently lower upload speeds lately ........

........ But my Netgear was pretty much always set at 1458 and I saw upload speeds of well over 300 kbps with it set at 1458.

More food for thought!

When I say reduce Les, I am talking something like;

1500 = around 383

1472/77  = 370 ish

1458 and below = will vary from 200/350

Mo

[Sebby]

Speedtest.net is a strange one. Sometimes it gives impossibly high results, and other times, results that are much lower than is the case. It seems to depend on your location and which server you use. I just tend to always find the Namesco tester consistently accurate.

[LesD]

1458 and below = will vary from 200/350

I see, so both the speedtest.net and BBMax figures could be right given that they were obtained an hour or so apart.

[Rik]

Could be, Les. All speed tests are suspect, since they use varying methodology to simulate real transfers. The BT test is the most accurate, after that I tend to look for a large file from a known good server, and use a stopwatch.

[MoHux]

I see, so both the speedtest.net and BBMax figures could be right given that they were obtained an hour or so apart.

IMHO Yes, Les, but I must tell you, I speak only from my own tests/observations.  I am no tech 'ed!!

Mo

[LesD]

The BT test is the most accurate

I would second that but BT in their arrogant wisdom choose not to do an upload test and do not time and date their result, which is a pity considering how long it can take to get one on many occasions!

[Rik]

Though the tests are logged so that ISPs can access them.