-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Hannu Kotipalo kirjoitti Sunnuntai 10. Marraskuuta 2002 14:49:
are speeded up a lot. For example, wzebra opening database readout :
Before : over 700 seconds After : 12-13 seconds
A new version, replacing the earlier. I tested this with a few applications, looks like for most of applications this has not much effect. However, wzebra proves that the effect can be enourmous on some. (And Simcity tellls that it can be negative on some applications :-(
Changelog: - - This patch speeds up applications reading files by small packets - - 4 static 512 byte buffers (can be defined for n * n buffers, see file.c) - - Buffer invalidation mechanism changed. With several buffers this *should* fix the reported (thans to Tony Lambregts) slowdown on SIMCITY2000. I haven't got it, cannot test. Please test and see -debugmsg +file, it can give a hint about needed buffers and their sizes.
diff is made by 'diff -urN wine-20021031/ wine-20021031-patched/ > patch.diff'
Hannu
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Sorry, this was meant to wine-patches. Appearantly (of course..) reply-to is forwarded here..
Hannu
A new version, replacing the earlier. I tested this with a few applications, looks like for most of applications this has not much effect. However, wzebra proves that the effect can be enourmous on some. (And Simcity tellls that it can be negative on some applications :-(
Changelog:
- This patch speeds up applications reading files by small packets
- 4 static 512 byte buffers (can be defined for n * n buffers, see file.c)
- Buffer invalidation mechanism changed. With several buffers this *should*
fix the reported (thans to Tony Lambregts) slowdown on SIMCITY2000. I haven't got it, cannot test. Please test and see -debugmsg +file, it can give a hint about needed buffers and their sizes.
just a dumb question: shouldn't we better (for synchronous unix reads) convert the fd into a FILE* (cache those FILEs for a few streams, or all of them ?) and let libc do the buffering on file reading/writing ?
A+
Eric Pouech eric.pouech@wanadoo.fr writes:
just a dumb question: shouldn't we better (for synchronous unix reads) convert the fd into a FILE* (cache those FILEs for a few streams, or all of them ?) and let libc do the buffering on file reading/writing ?
This won't work because a FILE* is process-local, while a handle can be shared across processes. Of course there is the same problem with doing the buffering ourselves in the client, so Hannu's solution won't work reliably either. To be reliable the buffering would have to be done in the server, but there's obviously no point in doing that.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Alexandre Julliard kirjoitti Keskiviikko 13. Marraskuuta 2002 02:00:
Eric Pouech eric.pouech@wanadoo.fr writes:
just a dumb question: shouldn't we better (for synchronous unix reads) convert the fd into a FILE* (cache those FILEs for a few streams, or all of them ?) and let libc do the buffering on file reading/writing ?
This won't work because a FILE* is process-local, while a handle can be shared across processes. Of course there is the same problem with doing the buffering ourselves in the client, so Hannu's solution won't work reliably either. To be reliable the buffering would have to be done in the server, but there's obviously no point in doing that.
Technically, it might be a problem. Practicaly, there are propably no application that would suffer from it. But you are right, this is not a first priority problem.
Hannu
-
Alexandre Julliard -
Eric Pouech -
Hannu Kotipalo