Subject: | |
From: | |
Reply To: | |
Date: | Wed, 28 Jun 1995 16:09:19 -0700 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Gavin writes:
> file created just for that purpose. If you are FLOCK/FUNLOCKing your shared
> data mapped file, performance is going to suck.
Yep. The bigger the file, the worse the performance hit.
From my viewpoint:
1) this is a bug in FUNLOCK ... it should *not* be posting pages to
disk;
2) it was probably put into FUNLOCK as a workaround to some other
problem in MPE (my assumption).
Similar performance problems occur when doing an FCONTROL to "complete"
the I/O (FCONTROL 2), although more "legitimately".
<PLUG>
Here's the output from our CSEQ Nugget, which has some hints/notes/warnings
on selected intrinsics:
CSEQ [1.22] - NUGGETS [K.2805.22] Demo version[ ]
(c) 1989 Software Research Northwest, Inc.
For help at the CSEQ prompt, enter: ?
CSEQ [nm]: fcontrol
Procedure FCONTROL (
filenum : int16 ; {R26}
controlcode : int16 ; {R25}
param : anyvar UInt16 ) {R23, R24}
{Address type = LongAddr}
{controlcodes (selected values) }
{ 1 general -> param }
{ 2 Complete all I/O (can be slow on XL) }
{ 3 read hardware status --> param }
{ 4 set timeout interval (param = seconds) }
...
CSEQ [nm]: funlock
Procedure FUNLOCK (
filenum : int16 ) {R26}
{ CCE: ok }
{ CCG: denied, you didn't have the file locked }
{ CCL: error }
{NOTE: FUNLOCK has serious performance problems }
{on MPE/iX, scaling with the size of the file, }
{because it "posts" all pages of the file to }
{disk each time you call it. }
</PLUG>
--
Stan Sieler [log in to unmask]
http://www.allegro.com/sieler.html
|
|
|