Linux bind / dhcp iowait status

[Glenn Satchell]

I'd say that's on the high side. I run failover on a lowend Sun Netra
t1 with 256MB ram, 550MHz sparc cpu, 3000PCs and 4500 phones. Never
goes above a couple of percent cpu, and io wait is negligible.

How fast does your dhcpd.leases file grow? eg

ls -l dhcpd.leases; sleep 60; ls -l dhcpd.leases

and calculate the size difference from the listings.

2000pc with lease of four hours typically means they try to renew when
the lease is 50% expired, ie every 2 hours. That would be a new lease
(2000 hosts/1800 seconds) about once per second. Couple of hundred
bytes in a lease, so that shouldn't be a real problem.

I doubt it is named as it will cache a lot of stuff in memory, and dns
updates should only be occurring if the pc's lease actually expires
(probably every night).

I'd go for longer leases, say at least 1 day. I use 1 week for phones
because I always want them to work, even if the dhcp server is down.

Is there anything else running on the box, syslog, backups perhaps?

regards,
-glenn

>Date: Wed, 15 Mar 2006 09:42:17 -0500
>From: "Tom Greaser" <tgreaser at hsc.wvu.edu>
>To: <dhcp-users at isc.org>
>Subject: Linux bind / dhcp iowait status
>
>have a beefy box that shows 18.40 % iowait at PEAK times
>and as low as 5.50 % during work hours..
>
>have 2000 phones lease set to 1 day
>
>2000 pc lease set to 4 hours..
>
>running ddns on this host ...
>
>just wanting to see if this is "normal"
>
>will track down to see if its named or dhcpd doing the bulk of IO wait..
>
>Its an Redhat AS 4 update 3 box with mirrored scsi Ultra 320 drives
>10000 rpm
>bind-9.2.4-7
>dhcp-3.0.1
>
>from free -h 
>buffer 137meg       cache 1627 meg
>
>thanks for any echos
>