3.1.1 Failover Speed (was: Re: 4.1.0a1 Failover sync speed)

[Michael Kaegler]

At 12:42 AM +1000 5/30/08, Glenn Satchell wrote:
>Yeah, but the initial sync has to copy over 286,000 leases. You don't
>normally hand aout 286k leases at the same time. Is the time similar on
>a subsequent restart once the initial sync has comleted or does it
>improve?

Not normally, but I have fielded 1000 DHCP requests per second before 
(when a IBM printer went crazy), making the sync time a respectable 5 
minutes.
Subsequent starts are fine, startup is on the order of 5-8 seconds.

>Have you tried snooping the network to see what traffic was being
>transmitted?

I suppose that should be the next move. Any idea of what to look for?

>  >This may have been true once, but we tested this: setup a subnet with
>>a "range 10.10.10.4 10.10.10.5" and assigned '.4 fixed. We joined two
>>machines, neither of which were the MAC assigned '.4, and no other
>>machines were on the subnet at all. The first was assigned .5, the
>>second just generated a 'no addresses available' error.
>
>Umm, not necessarily. dhcpd sends a ping before offering the address to
>make sure it is free. If it gets a response then it will abandon that
>address and not attempt to re-use it until it has no other available
>leases.

We included that in our test by having no other devices in the subnet 
(other than a router at .1; the DHCP server was out of the subnet 
with an ip-helper). The '.4 address would not have gotten a ping 
response, and so if the server was going to assign it, it would have.

>Best practise says that you should not have fixed-address devices
>inside your dynamic ranges.

Not an option at the moment, random addresses were allocated from all 
over the subnet in the old system. We'd have literally hundreds of 
pool statements (x140 subnets. Unworkable.)
But like I said, this is tested working. Despite what you're both 
saying, practical testing has shown that an address in a pool will 
not be handed out if there is a fixed-address statement for it.
-porkchop
-- 
Michael "Porkchop" Kaegler, Sr. Network Analyst
(845) 575-3061 Marist College, Poughkeepsie, NY