On Wed, Oct 23, 2002 at 07:35:50PM -0400, jerry ely wrote:
> Due to the 'speed-of-light' thing, the culprit for the performance issue is
> TCP/IP. I had the privelage of speaking with an IBM genius in networking and
> he provided an easy explanation. TCP/IP uses 'sliding window' for
> flowcontrol. The window size can be up to 64k and varies depending on
> network conditions, configuration and other thngs. Assuming the full window
> during bulk downloads, you can send or receive 64k of data in one direction
> without waiting for an ack. Doing a bit of math reveals that 64k bytes end
> up being enough to keep a t1 busy even if there are long link delays.
Actually, longer windows can be used now ("window scaling" extension,
which is supported by Linux, but you need to explicitly enable it in
/proc/sys/net/ipv4).
Window size affects your transfer speed, but the thing you can
absolutely not work around is the latency. There is simply no way
to get 'instant' response.
Mirek
--
Psyche-list mailing list
Psyche-list@redhat.com
https://listman.redhat.com/mailman/listinfo/psyche-list
