[net2-wg] 600 pps, high throughput

[Philip Levis]

On Mon, 2006-03-13 at 13:38 -0800, Omprakash Gnawali wrote:
> 
> > The number I quoted -- 600 pps -- was for small (14 bytes, including
> > headers) packets. The numbers are in the T2 tech report. The
> > measurements were intended to demonstrate the possible RX bandwidth. As
> 
> 
> This is 600*14*8/1024 = 65 kbps
> 
> Tony et al. from UCLA have managed to get up to 140 kbps using packet
> trains they describe in the following paper:
> http://netlab.cs.ucla.edu/wiki/files/TS2006WCNC.pdf
> 
> Any idea, what kind of throughput we can get with larger packet size
> but with the fixes you mentioned in your email?
> 

If you read the T2 paper, you can calculate the idle overhead of the
stack. It can send 616 14B (20B total) pps, or 380 40B (46B total) pps.
The latter has a bandwidth of 380 * 46 * 8 or 139 kbps, 121 of which is
data. I am sure that if you increased the packet size you could get
more, as you'd cut down on the idle transitioning time.

Note, however, that these numbers do not include MAC backoff: the
measurement was how quickly the stack can receive/send packets, and MAC
backoff quickly introduces an upper bound.

If the idle time per packet is I, then

616 * (I + 160/256k) = 1
380 * (I + 368/256k) = 1

so the idle time is ~1ms per packet, but it looks like it's not
independent of the packet size (you do have to read it out, after all).

I'd hypothesis that if you sent 250 byte packets, then you'd see ~100
packets per second, and so approximately 200kbps. But there is no
experimental data of any kind to support this hypothesis.

Phil