bind 9.1.1 problem

[Tim Jones]

Hi bind-users ,

 i've just build bind 9.1.1 for Solaris 2.6 running on a E250 - 2 x 300Mhz
 1024Gb memory.

 this server is a cache only recursive server. When under heavy load the server starts
 dropping loads of dns requests and the "udpInOverflows" value from "netstat -s"
 starts going through the roof.

 We also build bind 9.1.1 on solaris 7 on a E4500 - 8x333Mhz & 2048Gb Memory so
 see what would happen.

 Similar things happened under heavy load, the cpu utilisation never go above
 55% but "udpInOverFlows" went balistic again and we started getting hugh DNS
 response times, very often getting 3 x 5 sec timeouts followed by "server
 failed to respond".

 looking in /var/adm/messages we also started getting thousands of "ICMP -
 Network not reachable (Bad Port)" which I guess is where the client response
 socket had timeout by the tim the server got around to sending a response.

 At peak times both times these server appear to be serving about 1300
 queries/sec.

 Anybody got any ideas about this one.

 Also can anybody confirm whether bind 8.2.3 can run on multiple processors
 under Solaris, it seems that bind 8.2.3 can't spawn threads on to mutliple
 processors, is this correct. If not how do I make it use all processors a box
 has to offer, I've used the "-n" option on bind 9.1.1 to this affect but can't
 find the equivilent on bind 8.2.3.

-- 
Regards,
 Tim                          mailto:Tim.Jones at ntli.net

-----------------------------------------------------------
Tim Jones - Unix Internet Systems Engineer
Internet Technology Group
ntl Group Ltd, ntl House, Bartley Wood Business Park, Hook, Hampshire. RG27 9XA
Tel - 01256 752828
Fax - 01256 752811
e-Mail - Tim.Jones at ntli.net
Mobile - 07798 712967
alt. Fax - 0870 7062073
alt. e-Mail - Tim.Jones at Demilink.net