Cell Relay Archive

[Paul Koning <pkoning@xedia.com>]

Srikrishna wrote:
> ...
> However, both PPD and EPD require VC queueing implementation in the switch I
> presume?

No -- neither does.  You need a few bits of state per VC to keep track
of the discard state (two bits if I remember right).  And you probably
want per VC tracking of the number of pending cells, but that's far
easier
than per-VC queueing.

> Having said that EPD is more efficient than PPD, note that it is only within
> that switch. If a subsequent switch drops the entire packet, the endpoint TCP
> will retransmit the entire IP packet which makes the first switch look
> inefficient but not the subsequent switch whihc caused the retransmit.

Mostly true but not entirely.  In nearly all NICs, cells are DMA'd into
host memory as they are received.  So if a partial packet comes in, that
does take system bus resources.  Also, for some NICs the partial packet
is given to the driver, which has to clean it up -- which means that
you get at least part of the per-packet software hit.

	paul