Slap*CAT* and horrible performance?

[Brandon Hume - OpenLDAP list <hume-ol@Den.BOFH.Halifax.NS.Ca>]

I've got a strange problem here.  I have 2.1.21 up and running with about
50k entries.  Previously, performance was reasonable.  Today, however, search
performance went into the toilet, and I'm not able to find a cause.

I'm also in the process of setting up a replica; so, I try to export the
existing directory using slapcat.  It's taking hours to do so.  I'm running
under the assumption that the problems are related.

I'm using 2.1.21 on a Solaris 7 machine, using bdb backend, which I really
haven't done any specific tuning to, yet.  Slapcat seems to be able to 
export about 16K/s, out of an LDIF I expect to hit about 50M.  I assume which
fields I index shouldn't matter too much with slapcat.  The disk is not
saturated; I can copy the indexes using 'dd' in a matter of seconds.

Write activity with our setup is very rare.

It doesn't appear to matter whether I have slapd running or not, performance
is poor either way.  Trussing slapcat shows about one second between the
write() to the LDIF and a following poll() call.  slapcat -d -1 shows
"bursty" entry_decode () calls and little else.

Anyone have suggestions on where to look for the problem?