Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: [EFM] Necessity of DBA mechanisms ...

see comments in-line:

>    I agree DBA issue does not come in P2P case. But I was replying
> to a thread on the DBA issues in a P2MP case and i was summarizing
> the earlier comments on that thread. 

The need for DBA can only be established through realistic simulations under defined worst-case application loads, not by consensus opinion. Back-of-napkin calculations as you have shown below do not establish that there is a *requirement* for DBA, see below. 

> A Gbps PON split 32 way represents 30 Mbps per user. I claim that
> DBA makes sense in this case. 20% of that is 6 Mbps.

A 32-way split is an implementation assumption which was not stated previously. This is another case of our interpretations of consensus differing, since I would have used 16-way as the "consensus." I don't believe that persons deploying these higher split ratios are that concerned about providing huge bandwidth. They can't have it both ways. 

If we assume 20% improvement via DBA is really accurate, is 24Mb below some application loading threshold, whereas 30Mb is above? Even at 40% inefficient, it has not been proven that 18Mbps is inadequate to support real application loads without a DBA mechanism. The simple implementation case of no DBA and fixed slots has to be proven inadequate on real application loads before the added complexity of DBA should be considered. 

What if this proof shows that a 16-way split is handled fine by the simple TDMA scheme @ 1Gbps, but 32-way is not, or is borderline? For a given set of operating assumptions, there is some threshold split ratio where above it DBA becomes increasingly necessary, and below it no DBA is necessary. I assume you agree with this since the decreasing split ratio case degerates to P2P, which you agreed doesn't require DBA on the EFM link.

From this it follows that when holding split ratio constant at some arbitrary value, increasing total available bandwidth and tweeking slot size will again result in thresholds where DBA is either necessary or not. Larger split ratios at fixed slot time lead to longer cycle times across the endpoints, so large split ratio will also impact ability to support time-critical traffic like voice (but slot time is a factor also). This will again result in a threshold point of 'will support' versus wont.

Get 250 color business cards for FREE!