<div class="gmail_quote">On Wed, Dec 12, 2012 at 8:33 AM, Bernhard Kohl <span dir="ltr"><<a href="mailto:bernhard.kohl@nsn.com" target="_blank">bernhard.kohl@nsn.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
When (re)starting a couple of virtual machines (about 10) on the same host<br>
it frequently happens that some of them use the same DHCP transaction ID.<br>
<br>
This problem can be solved by combining the current timer ticks and the<br>
least significant bits of the hardware address(es) to get a better<br>
randomness for the seed.<br></blockquote></div><br>I ran into the same problem on a several thousand node Infiniband cluster recently. We had hundreds of nodes booting with duplicate IPs because of these PRNG-based XID collisions. I ended up reverting 12767d2202e620e32aef3fbdd2c4ad30c4e5ac22 to get back to MAC/GUID-based XIDs.<br>
<br>Your patch seems like the right way to fix it, but I probably won't have an opportunity to test at large scale again anytime soon.<br><br>Thanks,<br>-Marcus<br>