single-machine load balancing

sschmidt at my-deja.com sschmidt at my-deja.com
Wed Apr 5 14:57:47 UTC 2000


Strange question:  Does anyone have any experience running multiple
instances of named (bind 8.2.2-P3 is what I'm using) on a single SMP
machine?  I'm running into a bottleneck on one processor right now, as
the named process doesn't spawn clients for each query (reasonable).
Rather than assign additional machines to the task, I'd rather just run
multiple instances of named bound to different ports on the same
interface, thus utilizing the spare processor cycles I have on the
other processors in the box.

I've tried creating a separate config file for the "new" named process,
instructing it to use different directories, listen-on a different
port, use different channel names, and still end up with this type of
debug-on-hang:

ns_init(/etc/named2.conf)
Adding 64 template zones
evSetTimer(ctx 0x80d55f8, func 0x80607f4, uap 0, due
954950127.014978000, inter 3600.000000000)
evSetTimer(ctx 0x80d55f8, func 0x8071d40, uap 0, due
954950127.015074000, inter 3600.000000000)
evSetTimer(ctx 0x80d55f8, func 0x806d964, uap 0, due
954950127.015105000, inter 3600.000000000)
evSetTimer(ctx 0x80d55f8, func 0x8060860, uap 0, due
954950127.015128000, inter 3600.000000000)
evSetTimer(ctx 0x80d55f8, func 0x807adbc, uap 0x40116230, due
954946549.028716000, inter 0.000000000)
evSetTimer(ctx 0x80d55f8, func 0x807adbc, uap 0x401162d8, due
954946547.030199000, inter 0.000000000)
evSetTimer(ctx 0x80d55f8, func 0x807adbc, uap 0x401162f0, due
954946541.031483000, inter 0.000000000)
evSelectFD(ctx 0x80d55f8, fd 20, mask 0x1, func 0x805de4c, uap
0x40110488)

Any help is appreciated.

Steve Schmidt


Sent via Deja.com http://www.deja.com/
Before you buy.



More information about the bind-users mailing list