Cell Relay Archive

[ws@shore.net (William Stallings)]

In article <34F434B6.503A50FC@txc.com>, Kaustabh Duorah <duorah@txc.com> wrote:

> I cannot seem to find a reason for the difference in performance. If as
> you say, that PPD discards only half a PDU on average, the same should
> hold for EPD. Because, statistically, in EPD the switch should hit the
> threshold after half a PDU on average and so discard only half the PDU.
> For complete PDUs after congestion beginning, there is no difference. In
> both cases, full PDUs will be discarded.
> Where is the discrepancy?
> 
The difference is that EPD reacts earlier to congestion. With PPD, you
only begin dropping a packet after you are forced to drop a cell. As a
result, on average, you are forwarding half the cells of a discarded
packet, to be discarded at the destination, wasting the intermediate
network resources. With EPD, no unnecessary cells are forwarded.

Here's another simple-minded way to look at it. Suppose that the switch
that does the discarding is half-way between source and destination. When
the switch invokes PPD on a packet, the entire packet travels from source
to switch, and half the packet travels from switch to destination, for a
total "waste" of 1.5 "units". With EPD, the entire packet travels from
source to switch and all of its cells are discarded at the switch, for 1.0
units of waste. Offsetting the EPD advantage is the fact that occasionally
the switch will discard a packet under EPD that would have successfully
made it all the way through under PPD. But this offset is not enough to
erase the EPD performance advantage.

Bill

|                | Descriptions, errata sheets and discount order info |
| Bill Stallings |              for my current books and               |
|  ws@shore.net  |       information on my forthcoming books at        |
|                |              http://www.shore.net/~ws               |