Re: NT3.51 DHCP SERVER PROBLEM

["Michael J. Lewis" <[e-mail]>]

We saw something like this with early Windows 95 beta client and a 3.51 
server but at the time, were so new to DHCP that we never tracked down 
the cause.  Once we became more experienced with DHCP, Microsoft 
released the production version of Windows 95 which did not exhibit the 
same behavior and we were never able to replicate the problem.

We did learn that Microsoft 3.51 DHCP server gives out the least 
available address available as opposed to rotating through the address 
list.  Therefore, if an address becomes unusable but the server doesn't 
know it, the server will continue to try to give out this address and 
will effectively "hang" DHCP communications on the network (unless 
someone is lucky enough to put in a request while the offer for the bad 
address is outstanding.)  This is not a problem with Microsoft clients 
since they don't bother to send DHCPDECLINEs if duplicate addresses are 
detected.


Rajesh Saluja wrote:
> 
> I am testing my DHCP client implementation with Windows NT 3.51 DHCP server.
> I have implemented the client as per the lateset draft (i.e. dhcp 0.6 draft)
> but I am facing some problems in DHCPDECLINE phase. I am sending the
> following in the DHCPDECLINE message:
> ciaddr = 0.0.0.0
> requested IP address option = address to be declined
>    But the server is not accepting this DECLINE and it is again sending the
> same IP address in the next cycle.
>     I thought to change the packet format as per the RFC1541. So I tried
> by sending the declined address in the ciaddr. In this case, the first
> offered address is being marked BAD by the server but if the server offers a
> new address in the next cycle and it is also declined by the client, the
> server is not marking it a BAD address and keeps on offering the same IP
> address.
>    Has somebody else also experienced the same problem? Can somebody tell
> me how the NT3.51 server hasbeen implemented? Your help will be highly
> appreciated.
> 
> Best Regards:
> Rajesh Saluja
> email: [e-mail]