PGM: experminents with the IF toolset Marc Boyer LIAFA - Univ. Paris 7 - France PGM: experminents with the IF toolset – p.1/21
The protocol PGM: experminents with the IF toolset – p.2/21
PGM PGM = Pragmatic General Multicast Pragmatic = basic reliability, no membership Multicast = diffusion trough a tree NAK-oriented PGM: experminents with the IF toolset – p.3/21
Ideal behavior Sender Node Receiver Application ODATA(1) ODATA(1) ODATA(1) ODATA(2) ODATA(2) ODATA(2) ODATA(3) ODATA(3) ODATA(3) ODATA(4) ODATA(4) ODATA(4) PGM: experminents with the IF toolset – p.4/21
How lost are recovered Sender Node Receiver Application ODATA(1) ODATA(1) ODATA(1) ODATA(2) ODATA(3) ODATA(3) 2 3 NAK(2) NAK(2) NCF(2) NCF(2) ODATA(2) RDATA(2) RDATA(2) ODATA(3) PGM: experminents with the IF toolset – p.5/21
If the NAK is “too late” Sender Node Receiver Application ODATA(1) 1 ODATA(1) ODATA(1) ODATA(2) 1 2 ODATA(3) 1 2 3 ODATA(3) 2 3 ODATA(4) 2 3 4 ODATA(5) 2 3 4 3 4 5 ??? 2 3 4 5 PGM: experminents with the IF toolset – p.6/21
Loss unrecoverable Sender Node Receiver Application ODATA(1,1) 1 ODATA(1,1) ODATA(1) ODATA(2,1) 1 2 ODATA(3,1) 1 2 3 ODATA(3,1) 2 3 ODATA(4,2) 2 3 4 ODATA(5,3) 2 3 4 3 4 5 2 3 4 5 LOSS(2) ODATA(3) ODATA(4) ODATA(5) + SPM messages PGM: experminents with the IF toolset – p.7/21
First experiment: untimed model PGM: experminents with the IF toolset – p.8/21
Basic property Are all messages delivered or declared as lost ? topology S-N-R FIFO channels only ODATA can be lost SPM rate linked to the advance of the window PGM: experminents with the IF toolset – p.9/21
Last window The last window can be “frozen” Sender Node Receiver Application ODATA(6,4) 4 5 6 ODATA(6,4) ODATA(1) ODATA(7,5) 5 6 7 ODATA(8,6) 6 7 8 ODATA(8,6) 7 8 ODATA(9,7) 7 8 9 7 8 9 Waiting... add of a “closing SPM” PGM: experminents with the IF toolset – p.10/21
Closing SPM Sender Node Receiver Application ODATA(6,4) 4 5 6 ODATA(6,4) ODATA(1) ODATA(7,5) 5 6 7 ODATA(8,6) 6 7 8 ODATA(8,6) 7 8 ODATA(9,7) 7 8 9 SPM(9,10) 7 8 9 LOSS(7) ODATA(8) ODATA(9) PGM: experminents with the IF toolset – p.11/21
� Topology reduction Can we remove the nodes? Are both systems equivalent? Sender Node Receiver Sender Receiver PGM: experminents with the IF toolset – p.12/21
Equivalence point of view as a black box from the sender point of view from the receiver point of view Sender Sender Sender Receiver Receiver Receiver PGM: experminents with the IF toolset – p.13/21
Equivalence Result No equivalence for the receiver... Tuning the hiding relation: hiding the NCF hiding all but NCF and NAK PGM: experminents with the IF toolset – p.14/21
Second experiment: timed model PGM: experminents with the IF toolset – p.15/21
The model topology S-N-R FIFO channels with delays only ODATA can be lost SPM rate linked to the advance of the window Sender Node Receiver PGM: experminents with the IF toolset – p.16/21
Recovering a loss ∆ ODATA/SPM RTT NAK Did the NAK arrives “in time”? PGM: experminents with the IF toolset – p.17/21
✠ � ✞ ✟ ✑ ✁ ✒ ✓ ✒ ✁ ✓ ✖ ✌ ✎ ✍ ✗ ✏ ✌ ✝ ✏ ☛ ☞ ✁ ✎ ✆ ✝ ✞ ✠ ☛ ✌ ✆ ✍ ✌ ✎ ☛ ☞ ✌ ✍ ✌ ☞ Duration ✂☎✄ ✟✡✠ LOSS_MAX PROD_RATE ✆✕✔ Window Size Mess seq. num. AMB_RATE PROD_RATE Note: false for the last messages ( LOSS_MAX Window Size ) PGM: experminents with the IF toolset – p.18/21
✂ ✆ ✏ ✎ ✄ ✖ � ✁ Lifetime of messages in transmit window nothing specified in the RFC a sender can drop each message once send our model: try to keep the max. numb. of packets constant rate fom producer Window Size PROD_RATE PGM: experminents with the IF toolset – p.19/21
☛ � ✁ � ✂ ☛ ✝ ✞ ✟ ✠ ✡ Experiments Recovering all losses, except first and last window: Window Size LOSS_MAX AMB_RATE PROD_RATE (1) ✄✆☎ see ProdRateVsLosses.pdf see WindowVsLosses.pdf see AmbRateVsLosses.pdf PGM: experminents with the IF toolset – p.20/21
Conclusion work under progress trying to generate the relation with TReX PGM: experminents with the IF toolset – p.21/21
Recommend
More recommend