[Nottingham] Odd behaviour in KDE

Lee nottingham at mailman.lug.org.uk
Thu Aug 21 14:39:00 2003


> My suggestion was based on a common NFS tweak where there are problems.  Main 
> idea was bringing the NFS packet size to down below the MTU.  If a packet is 
> dropped, then only that one needs retransmission, whilst with 8KiB size and 
> 1500 ethernet MTU, then 6 packets will be retransmitted.
> 

Really? you'd think nfs would just request the 'missing' packet rather
than the whole segment? I'm sure this does'nt happen with tcp, but I
nfs/udp is built for throughput rather than error recovery, at least
there's lot of cool option to tweak with nfs....god I know why I hate
smb so much now....

Interesting stuff, aint nfs just so much smarter than smb, it would be
great if nfs had better security.. .... hahaha ;-). I'm never touching
smb again.....oh...unless somebody pays me...


> I think Iain should actually still consider applying it, once things settle 
> down, if he uses NFS much at all, as a performance optimisation.
> 

I'd agree there, retransmitting an 8k block on wireless is not a good
idea for performance..

> TCP/IP doesn't work very well if you have significant packet loss, that kills 
> throughput, with 50% it falls to 0.  Even with 20% you will spend all your 
> time backing off, and gradually trying to increase bandwidth consumption.

yeah, it's a shame tcp 'sees' all packet loss as congestion rather than
an packet loss caused by the linklayer... :-(, wireless tcp anyone?
Packet loss during the inital tcp 3 way handshake could be real bad
too.......

> 
> NFS likes a cleaner connection, because each transaction 8KiB maps to 6 UDP/IP 
> packets, and as the application is responsible for retransmission, it can 
> only retry the whole block.
> 

:-(((.... It's reminds me of atm cell loss, loss one atm cell , and you
have to retransmitt the whole aal segment again..


> If you look at SNMP  you'll see it's designed to be robust in face of much 
> less clean networks than TCP/IP would require.  NTP servers will not care if 
> 3/4 of the packets are dropped either, just so long as some of them get 
> through, they can always reduce the polling period.
> 

SNMP...yeah, it's great...but again..the security sucks the big one.
;-).

So packet loss caused by congestion is okay, packet loss through link
layer error's can be real bad... 

What we need is wireless tcp... perhaps, and get away for this god awful
layer 2 retransmittion system than 802.11b has... it's like shooting
flea's with a machine gun..... 


it's only when you really look a tcp this close, that you find it really
is not that clever..but it works....we just need bigger pipes, and less
people using microwave oven's....hahaha

> Rob
> 
> 

Laters,
Lee