Error "ber_get_next on fd 11 failed errno=11 (Resource temporarily unavailable)" and connection failure while testing and after 1-2 minutes.

[Сергей Богомолов <sereja84@mail.ru>]

Hello!

We recently installed OpenLDAP 2.3.24 on our Intel Xeon 3.8 GHz server.
My task - test OpenLDAP performance.
I use tools: SLAMD 1.8.2 (www.slamd.com), and DirectoryMark 1.3
(www.mindcraft.com/directorymark/)

For testing with DirectoryMark 1.3 I include directive to slamd.conf:

allow bind_v2 bind_anon_dn

(else was error : Error no protocol)
Not test starts, but after 1-2 minutes client system shows couple of error
messages (error! Can't contact LDAP server)
I try to 'telnet' server (telnet port) - it's okay.
Remote telnet to LDAP port - failed.
Local telnet to LDAP port - okay!
Server does not accept requests
In logs:

connection_get(11): got connid=1
connection_read(11): checking for input on id=1
ber_get_next
ber_get_next: tag 0x30 len 5 contents:
ber_get_next
do_unbind
ber_get_next on fd 11 failed errno=11 (Resource temporarily unavailable)
connection_closing: readying conn=1 sd=11 for close
connection_resched: reaquiring locks conn=1 sd=11
connection_get(11): got connid=1
connection_read(11): closing, ignoring input for id=1
connection_resched: attempting closing conn=1 sd=11
connection_close: conn=1 sd=11

After sleeping 2-3 minutes server is able to response again. (Rebooting
server also let it's response again).

Base=10.000 users, no replication.

Second I try authentication test on SLAMD 1.8.2 (AuthRate). 
(User connect to server, and, if successfully, disconnects)
Single user test shows very good performance (on 5 clents and 10 threads
more than successful 3000 auths/second). 
Random user logins (1009 base) test show some error logins even on 1 client
with 5 threads. (about 11 fails within 5 minutes (300.000 sucesses). 
Base = 10.000 users.

OS: Solaris 8 X86 02/02 
CPU: Intel Xeon 2.8 GHz
RAM: 1Gb (OpenLDAP "eats" 400 mb)
HDD: 50 Gb free

How to prevent server from "sleep"?

Thanks,
Sergey