What is the usual MTU on the Internet

[euston]

I would like to know what is the usual size of MTU on the Internet. Also, is it OK to size MSS as MTU minus 40? N.B. We are not a military establishment!

 

[jimbopalmer]

The usual MTU of serial ports is 576, of broadband/ethernet is 1500.

http://www.dslreports.com

has a tweak tool to see if yours is set right I tried to remain child-like, all I acheived was childish.

 

[euston]

Thanks jimbopalmer. I have some information for you which may or may not be useful.
I use ADSL in the UK. The infrastructure operator (British Telecom) suggests a MTU of 1458 to avoid packet fragmentation. I achieve an adequate RWIN by multiplying (MTU-40) by an arbitrary factor, say 14. The result is quite good performance; not far below theoretical in fact. I have only just found that Win2000 provides the facility of the Van Jacobson retransmit/recovery algorithm. I imagine that this is very effective in helping recovery from packet loss, especially with larger RWIN values. However, I don't seem to get much packet loss so haven't been able to test that assertion properly.

 

[jimbopalmer]

this is just a guess, but it may be right, if your user packets are futher encapselated (PPPoE or PPPoA come to mind) they may need to lower your MTU slightly so the final packet size does not exceed 1500 even including the wrappers put on it before it reaches the net

as I say, i have no proof but it could be why it is just under 1500 for you and 1500 for me I tried to remain child-like, all I acheived was childish.

 

[jgercken]

Why not use MTU Discovery? In Windows,

http://www.winguides.com/registry/display.php/888/

Note: setting this key to 0 will cause windows to use a max MTU of 576 bytes with all connections to other subnets.

I can't see why you can't set the MSS to MTU-40 except that there are options that can increase the header sizes over 20bytes (EG SACKS and timestamps). Sounds safe enough.

The biggest performance adjustment you can make is with the windowing. The Internet is high bandwidth, high delay so even though you may have the bandwidth the ack's delay keep the xfer rates down. Maybe enable window scaling?
-Jeff
----------------------------------------
Wasabi Pop Tarts! Write Kellogs today!

 

[NickFerrar]

speedtest.btopenworld.com/mtu

Has a utility to download to set your MTU to 1458, if you run it a second time it resets everything back to what it was - easier than registry editing.

 

[euston]

Thanks Nick. I have had a look at that web page. I assume that the figure of 1458 is down to the routers that BT uses.
I wonder what MTU is used with ADSL in other countries than the UK.

 

[euston]

For what it's worth, these are the TCP/IP settings I use for ADSL (512Mbps) : -

MTU = 1458 (recommended by UK telco Britsh Telecom - my ADSL carrier)
Using IP header value = 40 (consistent with both Time Stamping and SACKs = Off),
MSS = 1418 = (MTU - IP header value)

RWIN = 19852 = MSS*14 (arbitrary constant)
Windows Scaling = Off
TTL = 128
Path MTU Discovery = On
Black Hole Detection = On

Van Jacobson retransmit/recovery algorithm = On (Win2K/XP feature, default = Off)
I have a feeling that this last feature is rather useful if packet loss occurs.

N.B. The RWIN value is one which is optimal for 300ms latency.

ADSL just flies!

 

[euston]

Regarding Van Jacobson, I became aware of this tweak in Jermar Software's 'Tweaki...for Power Users' (version 3).
In the Win2000 section under the network tab is a checkbox to tick to enable this feature.

 

[euston]

FOR UK ADSL USERS

These are my Windows 2000 registry entries for maximum TCP/IP performance on the Internet (should be same as XP): -

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters]
"DefaultTTL"=dword:00000080
"TcpRecvSegmentSize"=dword:00000572
"TcpWindowSize"=dword:00005720
"EnablePMTUDiscovery"=dword:00000001
"EnablePMTUBHDetect"=dword:00000001
"Tcp1323Opts"=dword:00000002
"SackOpts"=dword:00000001
"TcpMaxDupAcks"=dword:00000002

I recommend a good read of RFC1323.
The above settings enable the Van Jacobson retransmit/recovery algorithm.

N.B. Anyone who uses these settings does so at their own risk. Do take care to make sure that they are suitable for your own purposes. I have found that setting Window Scaling to ON packets is a NoNo - a 16% performance loss results. I also observe that the

http://www.adslguide.org.uk

speed test is only a rough guide to actual speed. My settings do not yield the best statistical performance but do yield the fastest PERCEIVED performance and that is what matters to me.