home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.modems
- Path: sparky!uunet!utcsri!geac!torsqnt!uunorth!anton
- From: anton@analsyn.gts.org (Anton J Aylward)
- Subject: Re: UUCP 'G' protocol packet size
- Message-ID: <1992Sep6.204014.8017@analsyn.gts.org>
- Organization: ASCI: UNIX Database and Communications
- X-Newsreader: Tin 1.1 PL4
- References: <1079@bazooka.amb.org>
- Date: Sun, 6 Sep 1992 20:40:14 GMT
- Lines: 29
-
- ti@bazooka.amb.org (Ti Kan) writes:
- : I have been playing around with the UUCP 'G' protocol by increasing
- : my uucico's supported window size from 3 to 7. This yielded a nice
- : ~20% boost in throughput when transferring compressed news batches.
- : This appears to have no ill effects as far as compatibility with other
- : systems go.
- :
- : ....
- :
- : With the previous paragraph in mind, I looked into the version of
- : uucico that I'm using and found that the "packet size" is hardwired
- : to 64 bytes. So the question is, is there any benefit in increasing
- : this to something like 4096 bytes? Do I run the risk of making my
- : uucico incompatible with other systems?
-
- Think about it for a moment.
- Assuming you don't have the WB bug...
- If the remote site doesn't like 7,4096 it will negotiate what it wants.
- The "paths" (or parameter in SVR4) just sets a starting point.
-
- Benefit? fewer system calls. Better response because of less
- processing switching, less overhead (but not dramitcally so) in
- calculating csum, less moving of data. Write out a simple simulation of
- the basic flow and you'll see what I mean. The big saving is in fewer
- system calls. Conventional wisdom: system calls are more expensive
- than subroutine calls within a process.
-
- --
- Anton J Aylward
-