| View previous topic :: View next topic |
| Author |
Message |
Eric Grange
Guest
|
 Posted: Wed May 04, 2005 9:17 am Post subject: New replay benchmark : XML Parser |
 |
|
Placed temporarily on
http://glscene.sourceforge.net/misc/MMUsage.7z
It's a 3.9 MB 7Zip archive which uncompresses to about 75 MB,
can't leave it for long at this url, so if it could be uploaded
to the fastcode site that would be nice, let me know when I can
delete it :)
It's a record from our XML parser opening an 11 MB XML file
(an Open-Office spreadsheet) and building a complete DOM
(the spreadsheet is then altered in memory and saved back).
Theoretically it should be characterized by the allocation of a
whole lot of rather small blocks. It's a single-thread bench.
It's possible we may be reaching the limits of the log replay,
as it takes less than 800 ms of runtime in the application with
old RecyclerMM, while the best score I got in the B&V (assuming
I didn't mess up the RepeatCount) were around 1000 ms for FastMM3
and BucketMem (1200 for RecyclerMM)
Not sure if it's the file loading time (is that part of the bench time?)
or if having to read all the log data affects the cache in some
negative fashion.
Also in the B&V, memory usage seems a bit higher (reports around 140 MB,
while the application stands at 95 MB according to TaskManager when
the DOM has been loaded)
Eric
|
|
| Back to top |
|
 |
Dennis
Guest
|
| Posted: Wed May 04, 2005 9:23 am Post subject: Re: New replay benchmark : XML Parser |
|
|
Hi Eric
Zip please ;-)
Dennis
|
|
| Back to top |
|
|
Eric Grange
Guest
|
| Posted: Wed May 04, 2005 9:34 am Post subject: Re: New replay benchmark : XML Parser |
|
|
| Quote: | Zip please 
|
http://www.7-zip.org/
And forget Zip ever existed ;)
The zip version 4 times larger, at a whooping 15.5 MB!
Eric
|
|
| Back to top |
|
|
miab
Guest
|
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Wed May 04, 2005 11:13 am Post subject: Re: New replay benchmark : XML Parser |
|
|
Hi Eric
Do you deny giving me a zip ?
Regards
Dennis
|
|
| Back to top |
|
|
Robert Houdart
Guest
|
| Posted: Wed May 04, 2005 11:25 am Post subject: Re: New replay benchmark : XML Parser |
|
|
Eric,
Very interesting benchmark, one does get some noticeable speed and memory
differences.
BucketMM;1088;141156
FastMM2;1910;134952
FastMM3;1015;130844
NexusMM;5094;137528
PSDMM;1818;135204;
RecyclerMM;1073;140520
RTLMM;3071;128780
Even in this extreme case, with an average allocation size of only 41 bytes
and no memory recycling at all, the difference between the best and the
worst amounts to no more than 10%.
I suggest we include this benchmark in the B&V0.35 tool.
BTW, I agree that there's a rather constant 30 MB overhead in all the replay
benchmarks, we must have forgotten to take something into account ;-)
Regards,
Robert
"Eric Grange" <egrangeNO (AT) SPAMglscene (DOT) org> wrote
| Quote: | Placed temporarily on
http://glscene.sourceforge.net/misc/MMUsage.7z
It's possible we may be reaching the limits of the log replay,
as it takes less than 800 ms of runtime in the application with
old RecyclerMM, while the best score I got in the B&V (assuming
I didn't mess up the RepeatCount) were around 1000 ms for FastMM3
and BucketMem (1200 for RecyclerMM)
Not sure if it's the file loading time (is that part of the bench time?)
or if having to read all the log data affects the cache in some
negative fashion.
Also in the B&V, memory usage seems a bit higher (reports around 140 MB,
while the application stands at 95 MB according to TaskManager when
the DOM has been loaded)
Eric
|
|
|
| Back to top |
|
|
Pierre le Riche
Guest
|
| Posted: Wed May 04, 2005 11:26 am Post subject: Re: New replay benchmark : XML Parser |
|
|
| Quote: | And forget Zip ever existed
|
Zip? What's that? <g>
|
|
| Back to top |
|
|
Pierre le Riche
Guest
|
| Posted: Wed May 04, 2005 11:35 am Post subject: Re: New replay benchmark : XML Parser |
|
|
Hi Robert,
| Quote: | BTW, I agree that there's a rather constant 30 MB overhead in all the
replay benchmarks, we must have forgotten to take something into account
|
The replay file is loaded into memory and gets added to the usage.
A while back I was concerned when I noticed that the memory usage growed
slightly if I ran a benchmark many times in succession. Eventually I
realised it was not because of my MM fragmenting, but because the B&V tool
was caching the results in memory.
Maybe we should add functionality to the TBenchmark class to subtract the
inital VM usage (before the benchmark started) from all usage measurements.
But then again, that would advantage MMs that don't free VM back to the
OS...
Maybe we could allow each benchmark to subtract a fixed amount from its
usage?
Regards,
Pierre
|
|
| Back to top |
|
|
Eric Grange
Guest
|
| Posted: Wed May 04, 2005 12:42 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
| Quote: | Do you deny giving me a zip ?
|
If you have an FTP where I can upload it...
It's too big for the attachments newsgroups, sending by mail,
or for the SF webspace, and not really suitable for the SF
File Release System (no need to have it mirrored all around
the world).
Eric
|
|
| Back to top |
|
|
Eric Grange
Guest
|
| Posted: Wed May 04, 2005 12:48 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
| Quote: | The replay file is loaded into memory and gets added to the usage.
|
That's what I thought, but shouldn't that result in something closer
to a 70 MB difference in this case?
The application I recorded from doesn't use 40 MB for the rest of
its stuff - or if it does, it does it behind my back! ;)
Eric
|
|
| Back to top |
|
|
Robert Houdart
Guest
|
| Posted: Wed May 04, 2005 12:55 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
With the Replay Log Analysis Tool you get a 102 MB peak memory usage with
the "16-byte aligned" option. With the option "+4 and 16-byte" you get 113
MB peak memory.
All this is obtained by simply summing up all the memory requests.
The values match quite well the 130 to 140 MB usage reported by the B&V tool
if you take into account the +- 30 MB overhead present in all replay
benchmarks. So it looks like the size of the replay log file is correctly
taken into account by the B&V.
Robert
"Eric Grange" <egrangeNO (AT) SPAMglscene (DOT) org> wrote
| Quote: | The replay file is loaded into memory and gets added to the usage.
That's what I thought, but shouldn't that result in something closer
to a 70 MB difference in this case?
The application I recorded from doesn't use 40 MB for the rest of
its stuff - or if it does, it does it behind my back! ;)
Eric
|
|
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Wed May 04, 2005 5:35 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
Hi Eric
Just mail it to me.
Regards
Dennis
--
Jeg beskyttes af den gratis SPAMfighter til privatbrugere.
Den har indtil videre sparet mig for at få 86 spam-mails.
Betalende brugere får ikke denne besked i deres e-mails.
Hent den gratis her: www.spamfighter.dk
|
|
| Back to top |
|
|
Eric Grange
Guest
|
| Posted: Mon May 09, 2005 12:28 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
| Quote: | Just mail it to me.
|
Too bit for the outbox here... (rejected by the ISP).
Eric
|
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Mon May 09, 2005 12:49 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
Hi Eric
I can just download from your site?
My inbox is not to big either.
Regards
Dennis
|
|
| Back to top |
|
|
Eric Grange
Guest
|
| Posted: Mon May 09, 2005 1:17 pm Post subject: Re: New replay benchmark : XML Parser |
|
|
| Quote: | I can just download from your site?
|
File is too big for uploading to SF.Net webspace, and I don't think it's
really clean to upload it to the file release system (which would end up
duplicating it on all SF mirrors around the world).
Eric
|
|
| Back to top |
|
|
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
