[Date Prev][Date Next] [Chronological] [Thread] [Top]

2.1.17 performance issues

I've noticed that in 2.1.16 and 2.1.17, performance is pretty miserable,
and doesn't show the common signs of an application being beaten on. 
We're running on a dual sparc/450 with a 1GB of RAM under Solaris 8.  The
load average is consistently less than .1, slapd never consumes more than
5% of the cpu, and more often sticks around <1%.

By itself, our directory pushes about 170 queries/second (when performing
the same filter repeatedly), but during these "bogged-down" times, that
number drops to less than 100 queries/second and things back up very
badly.  The directory has only 5400 entries but is used fairly heavily by
postfix, courier-imap, and Oracle's CorporateTime.  What doesn't make any
sense, IMO, is that slapd doesn't necessarily consume a larger amount of
cpu time when more queries come in, and in fact, it's during the "bad
times" that it appears to be consuming *less*.  The load on the machine
never goes up, there's never any iowait, etc.

Are we looking at more threading issues here (I noticed one in the 2.1.17
changelog)?  I felt somewhat forced into the 2.1.16 upgrade once it was
marked as stable to workaround the buffer overflow issues in 2.0.?, and
I've run into enough issues along the way to wonder how ready for
primetime 2.1.x is...

Is anyone else experiencing such daunting performance problems?

OS: Solaris 8
LDAP: 2.1.17
DB: Berkeley 4.1.25
Hardware: Way more than I should need


John Madden
UNIX Systems Engineer
Ivy Tech State College