| 1 |
Rmst Filter README |
| 2 |
June 12, 2003 |
| 3 |
|
| 4 |
Fred Stann |
| 5 |
[email protected] |
| 6 |
|
| 7 |
The code in this directory constitutes the initial release of the Rmst |
| 8 |
Filter. The Rmst filter provides guranteed delivery and fragmentation |
| 9 |
/ reassembly of large blobs. Rmst has had limited deployment in an |
| 10 |
actual sensor network. To that extent the Rmst filter should be |
| 11 |
considered experimental. It has been "tuned" to the environment in |
| 12 |
which it was deployed (the ISI testbed). New environments may present |
| 13 |
traffic scenerios that cause Rmst to have sub-optimal behavior. The |
| 14 |
best way to receive help debugging such scenerios is to run the filter |
| 15 |
with the "-d 10" command line option and capture the DiffPrint screen |
| 16 |
output at key nodes (like sources and sinks), then email them to |
| 17 |
[email protected] along with a problem description file. |
| 18 |
|
| 19 |
Major Caveats: |
| 20 |
|
| 21 |
Rmst has only been tested in conjunction with the "two_phase_pull" |
| 22 |
routing algorithm in Diffusion. |
| 23 |
|
| 24 |
This initial release does not have ns-2 support. A release with ns-2 |
| 25 |
support should follow shortly. |
| 26 |
|
| 27 |
Rmst has only been tested with a single sink. |
| 28 |
|
| 29 |
This release is only intended to run on top of an ARQ MAC layer like |
| 30 |
S-MAC that provides hop-by-hop repair. Rmst "caching mode" support |
| 31 |
(in which each node caches fragments rather than just sources and |
| 32 |
sinks) is not available in this release. Future releases will |
| 33 |
re-enable this feature which was present in the simulation code used |
| 34 |
during research and development. |
| 35 |
|
| 36 |
Rmst has a coarse grained pacing mechanism for applications. After |
| 37 |
sources call sendRmst, they should not send another large blob until |
| 38 |
they get an RMST_CONT message from the filter. |
| 39 |
|
| 40 |
|
| 41 |
Users may need to manually adjust certain parameters to tailor Rmst to |
| 42 |
their particular environment. Down the road these parameters may |
| 43 |
become self adjusting or adaptive. The most important are enumerated |
| 44 |
here: |
| 45 |
|
| 46 |
1.) Refer to the Diffusion API and sample rmst application code in |
| 47 |
this release. You will note that the call to dr_->sendRmst has a |
| 48 |
third parameter called "fragment_size." Different sensor nets have |
| 49 |
different MTU limitations. Even MAC layers that do |
| 50 |
fragmentation/reassembly can only handle a modest sized payload. |
| 51 |
You must adjust fragment_size such that the combination of all your |
| 52 |
attributes, packet overhead, and payload do not exceed your network |
| 53 |
MTU. Rmst will break up any blob that is larger than the |
| 54 |
fragment_size into multiple fragments. |
| 55 |
|
| 56 |
2.) In rmst_filter.hh you will find some tunable parameters. Most |
| 57 |
notable is the "SEND_INTERVAL." This dictates how often packets are |
| 58 |
injected into the sensor network at the source. The fragment send |
| 59 |
rate is not controlled by the application when sending a large |
| 60 |
blob. Rmst must interleave NAK responses and new packets while |
| 61 |
pacing the overall load that it presents to the network. This rate |
| 62 |
could be made adaptive given the addition of some sort of |
| 63 |
backpressure mechanism to diffusion. |
| 64 |
|
| 65 |
3.) Rmst has a coarse grained blacklisting mechanism whereby links |
| 66 |
that deliver a low percentage of the packets placed on them are |
| 67 |
eliminated from contention during path discovery. This percentage is |
| 68 |
tunable in rmst_filter.hh. |
| 69 |
|
| 70 |
Thanks, |
| 71 |
Fred Stann |