How do I slam a DHCP server as hard as possible?

jcpunk · n00b Joined: 22 Jul 2003 Posts: 37

We have got an old server (Sparc5) with a 10BaseT network card in it that is currently our DHCP Server, and I have been assigned to prove that this 'new' system (P3 400) with a 100BaseT network card running Gentoo can provide addresses out under greater loads.

It is obvious to me that the 'new' system is so far superior to the old one (Processor, RAM, Network Card) that there should be no contest, but I cannot seem to write up a test program that will actually put load on the system.

My test lab is only 4 computers and my curent script cannot put enough requests out at at time to put any dent in either of these two machiene's processing power. A total of 12 per second is really not enough to have any impact. The problem is, I cannot use any existing dhcp program to test these with as the programs are deliberatly written not to allow these type of conditions.

My attempt looks like:

georwell · Posted: Fri Nov 05, 2004 9:36 pm Post subject:

Try adding hundreds or thousands of network interfaces to your network card. Then keep releasing, removing, taking up and down each interface in random.

However, I have never seen more then 20 ip addresses assigned to one card. So it might take a bit of work.

Just a thought!

NewBlackDak · Posted: Sat Nov 06, 2004 10:07 am Post subject:

That's a good idea.
Setup NTP, so the time in synchronized on all of them.
Setup ~20 interfaces on each machine. Write the script to take them all down at the same time, and bring them all back at the same time. Cron it to run on all of your machines at the same time. That would be 80 requests at once. If you have extra nics lying around drop those into the test machines, and do the same thing with those too.
_________________
Gentoo systems.
X2 4200+@2.6 - Athy
X2 3600+ - Myth
UltraSparc5 440 - sparcy

jklmnop · n00b Joined: 18 Jun 2003 Posts: 42

look at the man page for dhcpcd. there is a -T testing option which makes no changes to your running config. there is also a clientid option that you can set to random MAC addresses to force the server to give you a new lease.

nobspangle · Posted: Sat Nov 06, 2004 2:38 pm Post subject:

speed_bump · Posted: Sat Nov 06, 2004 3:21 pm Post subject:

Let's back the train up for a second. Is your network generating a significant load on the DHCP server at this time? DHCP server performance could possibly be an issue for very large ISPs, but it's typically not an issue even for healthy size networks (1000-2000 hosts). Realistically, either of those two systems should be able to provide more than adequate performance as a DHCP server (unless you're using them for other tasks as well).

There have been several discussions of this sort of thing on the DHCP discussion list which you can find over at ISC's web site (www.isc.org), so you may want to check the archives of that list for ideas.

NewBlackDak · Posted: Sat Nov 06, 2004 9:38 pm Post subject:

Our old DHCP(also DNS) was running on a sparc5 now that I think about it. It was an 85MHz machine with 256MB Ram. ~600 machines on 3 subnets, and the system monitor barely ever moved on it.
_________________
Gentoo systems.
X2 4200+@2.6 - Athy
X2 3600+ - Myth
UltraSparc5 440 - sparcy

speed_bump · Posted: Sat Nov 06, 2004 11:51 pm Post subject:

I've seen DNS run on some pretty lightweight machines - a P90 at one point (showing my age).

As you point out, there's no significant CPU load for this sort of thing.

In any case, for either DHCP or DNS you're more likely to run into memory issues long before network or CPU become an issue. In fact, in several discussions on the ISC DHCP list the indicate that you are more likely to see performance problems if there are a large number of leases moreso than if you're getting bombarded by a large number of simultaneous requests. This has to do with the data structures dhcpd uses to store the in-memory lease database.

There are other issues that may enter into it as well. In fact, the ICMP "ping before offer" can put a significant upper bound on performance. Obviously, it doesn't matter what architecture you're using if that's the case.

For my money I'd say either of those systems should be able to handle DHCP for all but the very most demanding environments. Unless there are actual specific performance problems that can be demonstrated, I think spending a lot of time benchmarking this would be largely pointless.

jcpunk · n00b Joined: 22 Jul 2003 Posts: 37

Here is the script I ended up with

screwloose · Posted: Wed Nov 17, 2004 9:14 pm Post subject:

What was the load diffrerence between your two machines with that script?
_________________
If something can go wrong it probably already has. You just don't know it yet. ~Henry's Modified version of Murphy's Law