I haven't seen any reply to this, so I'll take a swing at it.
Paul Wittry wrote:
> ---Reply to mail from Kirk Lawson about ppp speed
>>
>> Bad MRU/MTU settings?
>
> Speaking of MRU/MTU settings, what is the optimal settings for these
> with a linux box.
So long as the OS is decent, the OS has no impact on the optimal MTU/MRU
settings.
If your ISP has its act together, something called MTU negotiation
should completely obliviate your need to worry about this stuff. If
you're still interested in a much too long reply,...
Generally speaking, I set my MRU as high as reasonably feasible. For my
PPP connection, this is 9000. Will I ever get any packets close to that
size? No. But I don't want to take any chances on my host being the
one to reject a packet that I want because the packet just happens to be
a little large. (Yes, a packet that size would occupy my modem for
nearly three seconds; I'm fully aware of that.) This follows the
philosophy 'conservative in what you send, liberal in what you accept.'
MTU, on the other hand, should be dealt with more conservatively (as per
the philosophy.) *Most* of the time, if your kernel has not had MTU
discovery disabled, your system will figure out if your MTU is too high
for a particular site, and will lower it on a per-session basis. This
does mean that if your MTU is higher than your ISP's MRU, every TCP
connection through them will recieve some initial delay to start.
Frequent MTU settings are 1500 (Ethernet standard; an awful lot of the
internet is on this just because) and 576 (According to RFC 791, the
minimum MRU for IP networks, if I am reading it right. Anything on the
internet that isn't happy with this is therefore by definition broken,
unless it was made before January 1980 (release of RFC 760, which RFC
791 obsoletes), in which case it's merely obsolete.)
There has been one guy on the diald list (IIRC), who has the misfortune
to have to deal with a site that has their MTU set at 296, *and* is dumb
enough to block ICMP ping messages, so that MTU discovery can't work.
Oddly, he's wondered why that has problems with other software...
(Note to firewall people: if a machine has a reachable address, ICMP
ping smaller than or equal to the MRU should be able to reach it. It is
OK if you want to throttle the size of ping packets that get through, to
protect from ping of death. But dropping ping packets shorter than
about 9k (SMDS) is broken. Instead of dropping pings larger than the
MRU, reject them with destination unreachable, code 4.)
> How does one figure this out?
Trial and error with MTU discovery turned off.
> Where can I find relevent information on these subjects?
ftp://naic.nasa.gov/files/rfc?
Some of the HOWTOs mention some useful information.
*Sigh* I used to wonder about the people who spouted off RFC numbers in
response to a question... (In my defense, I tracked those RFCs down
just now; I hadn't memorized them.)
--
Ed
RFC1925 #8 "It is more complicated than you think."
-
To unsubscribe from this list: send the line "unsubscribe linux-diald" in
the body of a message to [EMAIL PROTECTED]
