[Dnsmasq-discuss] Estimation of TFTP Server Load Capabilities

Simon Kelley simon at thekelleys.org.uk
Thu Jul 31 20:07:24 BST 2014


On 24/07/14 04:13, Linux Luser wrote:
> I have a project where I use dnsmasq for netboot installs. Currently, there
> can be an unlimited number of installs happened at once. At what point
> (number of TFTP transfers happening in parallel) should I be concerned that
> I'm overtaxing dnsmasq's TFTP capabilities? Does dnsmasq use threads or
> multiprocessing for TFTP transfers?
> 

In my experience, the limit is unlikely to be dnsmasq, as long as it's
not running on a very slow CPU. The problem is that TFTP is a very dumb
protocol, and it really doesn't handle packet loss well. Nor does it
rate limit well. This means that the normal cause of collapse (ie hosts
never complete netboot) is network overload, causing packet loss,
causing TFTP sessions to stall. If you have a fast network, it may all
be fine, if you don't then changing the TFTP server probably won't help,
you need to limit the number of simultaneous transactions.

Cheers,

Simon.








More information about the Dnsmasq-discuss mailing list