On Sun, 24 Jun 2001, Steven James wrote:
> It may also be possable to speed things up by pre-requesting the next
> block and counting on gunzip to request it before the NIC's buffer can
> overflow with other traffic. For the first pass, I wanted to avoid any
> clever but potentially failure prone tricks. For a second pass, as
> long as the card is taken out of promiscuous mode, it should have no
> problems at all with that.
I think you made the right decision: keep it simple. As we add complexity
we will inevitably add failure modes, and in the worse case we'll start to
duplicate existing kernel capabilities. I'm still most comfortable relying
on using the kernel, though I completely understand the need to work
without one.
But the though of 128 clients trying to tftp is kind of interesting ...
ron
