[Dnsmasq-discuss] Patch: Add --ping-timeout option

[John Hanks]

On Tue, May 22, 2012 at 4:53 AM, Simon Kelley <simon at thekelleys.org.uk> wrote:
> On 21/05/12 16:56, Aaron Opfer wrote:
> For the installations where that doesn't work, namely large DHCP loads
> from netbooting many machines simultaneously. --no-ping is all that's
> needed.
>

FWIW I've used dnsmasq to netboot 1368 diskless machines
simultaneously (or as close to simultaneously as possible), which for
each machine meant pxe, tftp kernel/initrd, dhcp address with new
kernel, pull down image, switchroot to image, dhcp primary address
again plus 2 (or more) additional VLAN interfaces. In our case the
bottleneck was distributing the large OS image, but all 1368 nodes
(less a few hardware failures) could be booted in under 15 minutes
from power on to running compute jobs with the dnsmasq portions
fitting into the first 5 - 8 minutes of that 15 minute window. I never
used --no-ping and the only significant observed failures were a small
number of VLAN interfaces for which dhcp failed, which were addressed
by a startup script that waited a random number of seconds and
retried. At least in my experience DNSmasq was capable of handling our
load well. DNSmasq itself ran on a host with a 1 Gbps interface and
images were served from SSD on a host with a 10 Gbe interface. Of
course, YMMV.

jbh