home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cbmvax!andy
- From: andy@cbmvax.commodore.com (Andy Finkel)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: Clearing a PIPE:
- Message-ID: <34227@cbmvax.commodore.com>
- Date: 13 Aug 92 18:16:40 GMT
- References: <34141@cbmvax.commodore.com> <1992Aug11.124616.22374@pixel.kodak.com>
- Reply-To: andy@cbmvax.commodore.com (Andy Finkel)
- Organization: Flying Cat Software
- Lines: 57
-
- In article <1992Aug11.124616.22374@pixel.kodak.com> jeh@raster.kodak.com (Ed Hanway) writes:
- >andy@cbmvax.commodore.com (Andy Finkel) writes:
- >: Actually, the pipe-handler (actually queue-handler) in 2.0 was an entirely
- >: new implementation written by Fred Mitchell of the C= QA department.
- >
- >Any good reason for replacing Matt Dillon's perfectly adequate pipe-handler
- >with an inferior(*) replacement? It's not like C= can afford to reinvent
- >the wheel like this.
- >
-
- Hey, the previous poster was flaming the old pipe-handler to a crisp.
- He hated it. :-)
-
- But, as long as you ask:
-
- The pipe-handler turned out to the the main culprit behind hangs
- for the AS225 software. It apparently had some serious problems when
- multiple processes were banging on it (over the ethernet, anyway).
- (It was an early one from Matt) Anyway, Fred wanted to fix it, he had access
- to the old one's source, but decided its methods could be improved.
- (Matt himself used entirely different methods on his next pipe-like thing)
-
- >(*) Try pushing a megabyte through PIPE:. On a 68000 it took 90 seconds,
- >and that was with a single Write() call. Considering that's 50-100 times
- >slower than hard disk access, I was a little disappointed.
-
- The new one will allocate memory for the entire megabyte, unless a queue limit
- is specified in the mountlist. The old one started deferring the return
- of the write packets until the reader had read enough so there was
- enough space in the buffer again.
-
- If you are going to do large transfers, you might want to increase
- the blocksize that the handler uses, by putting a line like:
-
- BLOCKSIZE = 16384
-
- in your mountlist entry for queue-handler. (It defaults to 4096)
-
- While you can get the deferred reply effect from queue-handler as well,
- by specifying the number of buffers it is allowed to allocate.
- It defaults to 0, or allocate as many as it needs. Limiting it
- to a couple will give you the speed increase. But things are more robust
- if you let it allocate as many buffers as it needs. Especially when
- multiple processes over ethernet are involved :-)
-
- >Ed Hanway --- jeh@raster.kodak.com
-
-
- andy
- --
- andy finkel {uunet|rutgers|amiga}!cbmvax!andy
-
- "A new release is where old bad assumptions are replaced by new
- bad assumptions."
-
- Any expressed opinions are mine; but feel free to share.
- I disclaim all responsibilities, all shapes, all sizes, all colors.
-