| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 4180.1 | | WJG::GUINEAU | | Sun Oct 07 1990 11:09 | 31 |
| > If I boot up and start up a cli window (I use qmouse with its quickkeys)
> MIPS gives me 1.07 and disk speed now gives me a top of about 600kb/sec.
> If I then hammer the system with a few LHARC's, some heavy disk copies
> to ram: and then clean it all up so the disk and ram are exactly as
> they were and there is exactly the same CPU load etc the best I can
> get is about .84 MIPS and about 530-550 kb/sec. This ~10% change in
> performance is consistant and repeatable. Any ideas???
This almost sounds like a memory fragmentation phenomena, But I would
expect the memory to be coalesced into a big chunk once it was ass free again.
I've uploaded a program I wrote looong ago called show. It will display
a bunch of the Exec lists; one of them is the memory list, and more
importantly, the fragmentation. Run it before and after (show -m) and see
if there are any changes. (wjg::amiga:show.lzh)
You might also look at which tasks are running (show -t) and their priority
before and after.
> Another related issue - MIPS was run on a 2000 with a 28mhz GVP 030
> board and it gave 16MIPS. Similary my 68010 powered 2000 gives around
> the 1MIP area. Both these numbers seem to be higher than they should.
> What other figures does anyone else have and what do you all think of
> the MIPs output value?
16 mips?!? Wouldn't that be nice!
I have a 28mhz Imtronics 030 and it rates around 8.4 mips (using the mips
program). Something is definetly wierd here.
john
|
| 4180.2 | | LEDS::ACCIARDI | Probing the limits of adhesion | Mon Oct 08 1990 01:55 | 6 |
|
MIPS shows 4.0 in my A2620 equipped A2000.
I don't think this means very much.
Ed.
|
| 4180.3 | thanks - trying asap | MEO78B::MANDERSON | Photographers do it in darkrooms | Mon Oct 08 1990 06:43 | 8 |
| John,
I am pulling the show over now - will put a reply in with some info.
My guess was to do with memory fragmentation as well.
regards
k
|
| 4180.4 | | WJG::GUINEAU | | Mon Oct 08 1990 10:56 | 10 |
| I believe the Amiga exec will do "garbage collection" i.e. coalesce (sp?)
fragments into bigger hunks, however it probably only does this when
a memory allocation request cannot be satisfied because there is not a
large enough hunk on the free list - it then tries to combine all smaller
contiguous chunks into larger ones to satisfy the request. This will incur
some preformance hit and maybe that's why things are slower at first.
But I'd expect it to get better once the coalescing is complete...
john
|
| 4180.5 | some results so far | MEO78B::MANDERSON | Photographers do it in darkrooms | Tue Oct 09 1990 07:42 | 10 |
| Well after doing a lot of trials I am starting to wonder if what I am
seeing is a bug in MIPS. It seems to either be 'low' performance or
'high' performance. It does seem to relate more to 'low' after a heap
of work but my current tests are somewhat inconclusive.
I am going to keep trying different things to see if I can get a
real patter or if it relates to a specific exe somehow.
regards
k
|