Rob...
Here's a snipped version of a post I made some time back on the subject...
----------------
"Mama Don't Let Your Replay Files Grow Up to Be Monsters"
<snipped my drivel>
It seems that GPL doesn't quite like the idea of an 80MB replay file in a
128MB GPL server, and it had a rather unceremonious way of
letting us know it.
<more snipped>
With thanks to Randy [Cassidy of Papyrus], here's part of what he had to say
on the subject:
----
"From experiments I performed on Win95, it seems that Windows will only ever
allow a single running program to have as much as half the physical memory
in the computer. This includes the program, its data, and the pieces of
the operating system that it uses. Once a program tries to use more than
this, Windows seems to page other parts of that program (or its data) out
to disk, even if there are few other demands on memory....
Using a 64M replay buffer on a 128M system will be well beyond the point at
which Windows will swap out other parts of GPL as the replay grows. If
you're not noticing any significant performance degradation with a replay
buffer this large, then you'll probably not notice any with an even larger
replay buffer. (The parts of GPL that Windows will be swapping out will
most likely be old replay data anyway).
Be careful making the replay buffer extremely large, though. My machine at
work (PII-400, 128M) often crashed when I tried to save large replays
(>20M). If the replay is 64M, GPL 1.0 might ask Windows to write a single
64M block of memory to disk in one operation. Windows, or the BIOS, choked
very badly on this about 80% of the time, leading to a system crash (not
just a GPL crash). (The 1.1 patch will address this issue)."
----
So our experience now adds an underscore to Randy's cautionary comments.
We're now back to using 48MB replay buffers, despite earlier discussions
with others who reported success with replay files as large as 90MB. My
guess is that if you appear to be having success with a replay buffer that
large, you just haven't filled it up yet. You have to *host*, not join, a
long event, capture a lot of graphic effects, and have a large field before
you fill replay buffers of that magnitude.
Jack Rambo
Rob...
Yes. The amount of the replay memory allocation used is a function of, among
other things, race distance, the number of drivers (AI or otherwise) and the
number of graphics effects selected. The only difference I see between
online fields and comparably-sized AI fields is that the AI tend to produce
less in the form of smoke, fire, dirt, dust, sparks, and skid marks! :)
Jack Rambo
Thanks for the information, Jack. I don't/can't play on-line, so is this
information as relevant for me playing off-line?
--
Rob.
>Rob...
>Here's a snipped version of a post I made some time back on the subject...
>----------------
>"Mama Don't Let Your Replay Files Grow Up to Be Monsters"
><snipped my drivel>
>It seems that GPL doesn't quite like the idea of an 80MB replay file in a
>128MB GPL server, and it had a rather unceremonious way of
>letting us know it.
><more snipped>
>With thanks to Randy [Cassidy of Papyrus], here's part of what he had to
say
>on the subject:
>----
>"From experiments I performed on Win95, it seems that Windows will only
ever
>allow a single running program to have as much as half the physical memory
>in the computer. This includes the program, its data, and the pieces of
>the operating system that it uses. Once a program tries to use more than
>this, Windows seems to page other parts of that program (or its data) out
>to disk, even if there are few other demands on memory....
>Using a 64M replay buffer on a 128M system will be well beyond the point at
>which Windows will swap out other parts of GPL as the replay grows. If
>you're not noticing any significant performance degradation with a replay
>buffer this large, then you'll probably not notice any with an even larger
>replay buffer. (The parts of GPL that Windows will be swapping out will
>most likely be old replay data anyway).
>Be careful making the replay buffer extremely large, though. My machine at
>work (PII-400, 128M) often crashed when I tried to save large replays
>(>20M). If the replay is 64M, GPL 1.0 might ask Windows to write a single
>64M block of memory to disk in one operation. Windows, or the BIOS, choked
>very badly on this about 80% of the time, leading to a system crash (not
>just a GPL crash). (The 1.1 patch will address this issue)."
>----
>So our experience now adds an underscore to Randy's cautionary comments.
>We're now back to using 48MB replay buffers, despite earlier discussions
>with others who reported success with replay files as large as 90MB. My
>guess is that if you appear to be having success with a replay buffer that
>large, you just haven't filled it up yet. You have to *host*, not join, a
>long event, capture a lot of graphic effects, and have a large field before
>you fill replay buffers of that magnitude.
>Jack Rambo
Thanks for this information. I have been hosting great F3 online races, but
lately I have been getting more crashes then before I ran F3. Is the longer
race times a possible cause of system crashes? I have 128MB ram. I run the
races in 800x600, because I can, and I like the clarity.
Tore Hanson
>Rob...
>>...is this information as relevant for me playing off-line?
>Yes. The amount of the replay memory allocation used is a function of, among
>other things, race distance, the number of drivers (AI or otherwise) and the
>number of graphics effects selected. The only difference I see between
>online fields and comparably-sized AI fields is that the AI tend to produce
>less in the form of smoke, fire, dirt, dust, sparks, and skid marks! :)
>Jack Rambo
Tore...
Absolutely! Long races + large fields + huge replay buffer (sometimes) = GPL
crash.
If you're starting to fill a really large replay buffer, duck and cover.
Until 1.1 is released, I'm settling for 48MB on my 128MB server.
Jack Rambo
If you are running GPLength races, I suggest you don't let Windows
to control your power, like stop your HD. It's very *** in the
last lap, when leader crosses the finishline and suddenly your HD
wakes up and screen freezes for 5 seconds...happened to me in
Zandvoort...hitted the fence after the final corner at final lap...
Jack Rambo kirjoitti viestiss? ...
>Tore...
>>Is the longer race times a possible cause of system crashes?
>Absolutely! Long races + large fields + huge replay buffer (sometimes) = GPL
>crash.
>If you're starting to fill a really large replay buffer, duck and cover.
>Until 1.1 is released, I'm settling for 48MB on my 128MB server.
>Jack Rambo
