peter at rathlev.dk
Mon Feb 9 15:09:10 UTC 2015
On Mon, 2015-02-09 at 18:27 +0530, Vikas Jadhav wrote:
> *) dhcpd application support upto how much TPS
As Simon said this depends on what hardware you use and what other stuff
the server has to do apart from handing out leases. From our experience
the disk system IOPS is the limiting factor when using spindle disks.
We run our DHCP service in a failover setup on two IBM x3650 servers
each with a single Intel Xeon 5130 dual-core CPU and 4G RAM. The lease
file is placed on a RAM disk and our pools cover ~450k addresses all in
all. We handle ~175k requests per day. The servers aren't breaking a
When we have events where a lot of clients restart simultaneously the
load rises. We placed the leases file on a RAM disk to better survive
these events, since a regular spindle disk can often only handle around
150 IOPS and each handed out lease means at least one write. Take a look
at using SSDs to mitigate this, especially if you're not using a
failover setup that can survive one server crashing.
> *) How max IP pool ranges can able to add
RAM is the only real limit here. Service startup takes some time when
you have many leases, but standard server hardware can probably easily
handle 500k leases. Our dhcpd processes have a VSZ of around 320 Mbytes
with the aforementioned 450k pool addresses.
I don't think the actual number of pools as such have much bearing on
performance, just the total number of addresses in the pools.
More information about the dhcp-users