Dave Warren wrote:

> Chris Buechler wrote:
>
>> On Mon, 07 Mar 2005 17:31:00 -0700, Dave Warren
>> <maillist at devilsplayground dot net> wrote:
>>  
>>
>>> I've always been at least moderately annoyed that although my range
>>> .100-.199, only IPs in the .190-.199 range seem to get used, and those
>>> IPs get reassigned to new MAC addresses very quickly after the lease is
>>> given up.
>>>
>>>   
>>
>> That's how DHCP works on any platform, any server.  If you don't want
>> the IP's reassigned so quickly, use longer lease times.  It'll assign
>> the next available IP out of the range, and if a lease is expired that
>> IP is available.
>>  
>>
> My win32 DHCP server tends to start assigning IPs, but doesn't tend to 
> reassign IPs immediately after they expire unless there are no other 
> IPs available.
>
> I tend to use 7 day lease times, but it's not uncommon to have a 
> virtual machine stay offline for significantly longer then that.
>
> Is there a specific need to reissue the IPs immediately vs using the 
> entire range of IPs?
>

This is all observation only, and I don't know what the "official" 
behavior is supposed to be or claimed to be, but what I normally see on 
Windows DHCP servers at least, is that they appear to honor the client 
request for IP address when possible.  For Windows clients, they always 
request the same address they had last even if the lease has long since 
expired, assuming it's in the same subnet as where they found the DHCP 
server.  That essentially means that no matter what your lease time, a 
machine will get an IP address and keep it until the server is forced to 
issue it to someone else by virtue of running out of available addresses 
and having to start reusing them.  That could be colored by the fact 
that probably 90 percent of the machines on our network are never 
offline long enough for the lease to expire, but the remainder of them 
are off for weeks, even months at a time, and they always seem to wind 
up with the same address as well.  Our pool is substantially larger than 
the number of machines on the network at the moment thus there is never 
any reason to reissue an address to any other machine than the one which 
had it last.

-- 
Whom computers would destroy they must first drive mad.