Performance and memory pb with INN 2.3

Katsuhiro Kondou kondou at nec.co.jp
Wed Aug 2 16:32:41 UTC 2000


In article <20000802215225Z.kondou at inn.do.mms.mt.nec.co.jp>,
	Katsuhiro Kondou <kondou at nec.co.jp> wrote;

} } Yes, both memory usage and performance are the same.
} 
} Oops, I've misread your first mail.  You're running
} timecaf not tradspool.  I'll look into the code, but

Hm, timecaf has cache table internally in some condition when
retrieving artilcles(even when RETR_STAT).  The table is
created by hashing accepted time.  Possible table size of the
cache per class in storage.conf will grow to
sizeof(*ptr)*256*256*256=64MB and never release until exit.  I
thinks attached might help you, though still the performance
issue exists.  And as to performance, nnrpd needs to open/read
each crunched file.  This will consume disk i/o, since your
script behavior leads to retrieve all articles.
-- 
Katsuhiro Kondou

--- nnrpd//nnrpd.c.orig	Wed Aug  2 07:29:56 2000
+++ nnrpd//nnrpd.c	Thu Aug  3 01:29:16 2000
@@ -700,11 +700,6 @@
 #endif /* defined(DO_PYTHON) */
     
     val = TRUE;
-    if (SMsetup(SM_PREOPEN, (void *)&val) && !SMinit()) {
-	syslog(L_NOTICE, "cant initialize storage method, %s", SMerrorstr);
-	Reply("%d NNTP server unavailable. Try later.\r\n", NNTP_TEMPERR_VAL);
-	ExitWithStats(1, TRUE);
-    }
     if (!ARTreadschema()) {
 	Reply("%d NNTP server unavailable. Try later.\r\n", NNTP_TEMPERR_VAL);
 	ExitWithStats(1, TRUE);



More information about the inn-workers mailing list