Re: [Xen-users] MAC address management (and a guy with poor math skils talks about the birthday paradox!)

["Luke S. Crawford" <lsc@xxxxxxxxx>]

On Fri, 30 Mar 2007, Nico Kadel-Garcia wrote:
> Luke S. Crawford wrote:
> Assign each Xen *server* a MAC address range, and incorporate it into their 
> MAC addresses. Server 1 gets the last four digits of 01:00 thorugh 01:FF, 
> server two gets 02:00 through 02:FF, etc. That allows you plenty of addresses 
> for each Xen environment, and a meaningful way to look up hostnames of Xen 
> servers based on MAC address as well.
> Then hardcode both the MAC addresses and the vifnames accordingly, for any 
> SNMP resource tracking use.
Hm. those are good ideas.  (oh, and vifname= looks like it is going to 
solve a bunch of my problems at my other venture where we do properly 
hard-code the mac and IP.  Thanks!)    But like I said, this shop tracks 
IP addresses via ping.... when I brought this issue up, one of the  guys 
said "Can't you just make it check to see if the MAC address is taken 
before it brings up the interface?"  so something that requires less 
accounting is probably in order.
I'm considering changing vif-common to send out an 'arp who-has'  before 
grabbing an arp, and then having it back off (and trying another, if it's 
randomly generated)  if it's taken, much like the RHEL network scripts. 
are there any obvious problems with this approach.   It seems like it 
would work reasonably well if that change is accepted to the mainline xen 
kernel, and assuming that the mainline xen kernel uses a unique prefix.
(the places I see this not solving the duplicate mac address issue is 
in an 'xm save' situation, and when an older xen without the mac 
collision detection is brought up after a newer one on the same network.)
is this obviously a bad idea?

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users