How to handle massive dhcp request storm

Hi,

following situation:
I have a network with around 5000 dhcp clients. A situation can occur,
where all 5000 clients are switched off and on again at the same time.
Now, when all clients boot up simultaniously the dhcp server is not
able to handle that "request storm" and most of the clients get a
timeout and no ip.

Does anybody know a specifc dhcp server / server or client
configuration parameter which I can use to handle that problem, so that
all clients get an ip address as fast as possible.

Thanks & best regards

Marc

On 18 Oct 2006, in the Usenet newsgroup comp.os.linux.networking, in article
<(E-Mail Removed). com>, (E-Mail Removed)
wrote:

>I have a network with around 5000 dhcp clients.

What _possible_ reason could there be for that many clients using ONE
server? What happens when that one server goes tits up because a herd
of yaks just stampeded through your server room? If all those clients
were on a single network, that implies a netmask of 255.255.224.0
which is exceptionally bad traffic planning.

>A situation can occur, where all 5000 clients are switched off and on
>again at the same time. Now, when all clients boot up simultaniously the
>dhcp server is not able to handle that "request storm" and most of the
>clients get a timeout and no ip.

If you absolutely _must_ use dynamic addresses, install a LOT more DHCP
servers - so that this condition won't occur.

>Does anybody know a specifc dhcp server / server or client
>configuration parameter which I can use to handle that problem, so that
>all clients get an ip address as fast as possible.

While the original Ethernet specifications did allow for up to roughly
65000 hosts on a single collision domain, no one ever bothered because
the massive traffic jams that occurred. Our own network used a /22 mask
(255.255.252.0) which permitted 1022 hosts per subnet, though we never
had more than 700 per, and reconfigured it later with switches so that
no more than 50 hosts were on a given wire. I can't believe that
your network might be using a mask even that wide, and thus need DHCP
relay agents on your routers. You would make more sense to have a
DHCP server on each subnet - especially if the DHCP server has to
boot first before it starts handing out IP addresses.

Old guy

(E-Mail Removed) wrote:
> Hi,
>
> following situation:
> I have a network with around 5000 dhcp clients. A situation can occur,
> where all 5000 clients are switched off and on again at the same time.
> Now, when all clients boot up simultaniously the dhcp server is not
> able to handle that "request storm" and most of the clients get a
> timeout and no ip.

Are you sure ?

Since 5000 hosts on a single subnet are a network nightmare under any
circumstance, these will probably be routed over several *dozen* subnets
- yes ?

In this case, running a single DHCP server is... well, basically you're
waiting for the DHCP server to die, so you can work 24 hours straight
trying to solve the incident that will probably cost you your job.

Run. multiple. DHCP. servers.
Soon.
Like, now.

> Does anybody know a specifc dhcp server / server or client
> configuration parameter which I can use to handle that problem, so that
> all clients get an ip address as fast as possible.

Set up at least 2 or 3 DHCP servers with non-overlapping ranges in their
subnets, so every client has at least one other DHCP server to fall back
on when one goes down.

Configure your routers to properly relay DHCP requests from all clients.


J.