1. © 2000 Morgan Kaufman Overheads for Computers as Components Networks zNetwork-based design. yCommunication analysis. ySystem performance analysis. zInternet-enabled systems.

2. © 2000 Morgan Kaufman Overheads for Computers as Components Communication analysis zFirst, understand delay for single message. zDelay for multiple messages depends on: ynetwork protocol; ydevices on network.

3. © 2000 Morgan Kaufman Overheads for Computers as Components Message delay zAssume: ysingle message; yno contention. zDelay: yt m = t x + t n + t r y = xmtr overhead + network xmit time + rcvr overhead

4. © 2000 Morgan Kaufman Overheads for Computers as Components Example: I 2 C message delay zNetwork transmission time dominates. zAssume 100 kbits/sec, one 8-bit byte. zNumber of bits in packet: yn packet = start + address + data + stop y = 1 + 8 + 8 + 1 = 18 bits zTime required to transmit: 1.8 x 10 -4 sec. z20 instructions on 8 MHz controller adds 2.5 x 10 -6 delay on xmtr, rcvr.

5. © 2000 Morgan Kaufman Overheads for Computers as Components Multiple messages zIf messages can interfere with each other, analysis is more complex. zModel total message delay: yt y = t d + t m y = wait time for network + message delay

6. © 2000 Morgan Kaufman Overheads for Computers as Components Arbitration and delay zFixed-priority arbitration introduces unbounded delay for all but highest- priority device. yUnless higher-priority devices are known to have limited rates that allow lower devices to transmit. zRound-robin arbitration introduces bounded delay proportional to N.

7. © 2000 Morgan Kaufman Overheads for Computers as Components Example: adjusting messages to reduce delay zTask graph:z Network: P1P2 P3 d1 d2 M1M2M3 allocation 3 4 3 execution time Transmission time = 4

8. © 2000 Morgan Kaufman Overheads for Computers as Components Initial schedule time M1 M2 M3 network 0 2010515 P1 P2 d1 P3 Time = 15

9. © 2000 Morgan Kaufman Overheads for Computers as Components New design zModify P3: yreads one packet of d1, one packet of d2 ycomputes partial result ycontinues to next packet

10. © 2000 Morgan Kaufman Overheads for Computers as Components New schedule time M1 M2 M3 network 0 2010515 P1 P2 d1 P3 d2d1 P3 d2d1 P3 d2d1 P3 d2 Time = 12

11. © 2000 Morgan Kaufman Overheads for Computers as Components Further complications zAcknowledgment time. zTransmission errors.

12. © 2000 Morgan Kaufman Overheads for Computers as Components Priority inversion in networks zIn many networks, a packet cannot be interrupted. zResult is priority inversion: ylow-priority message holds up higher-priority message. zDoesn’t cause deadlock, but can slow down important communications.

13. © 2000 Morgan Kaufman Overheads for Computers as Components Multihop networks zIn multihop networks, one node receives message, then retransmits to destination (or intermediate). ABC hop 1 hop 2 Network 1Network 2

14. © 2000 Morgan Kaufman Overheads for Computers as Components System performance analysis zSystem analysis is difficult in general. ymultiprocessor performance analysis is hard; ycommunication performance analysis is hard. zSimple example: uncertainty in P1 finish time -> uncertainty in P2 start time. P1P2

15. © 2000 Morgan Kaufman Overheads for Computers as Components Analysis challenges zP2 and P3 can delay each other, even though they are in separate tasks. zDelays in P1 propagate to P2, then P3, then to P4. P2 P3 P1 P4

16. © 2000 Morgan Kaufman Overheads for Computers as Components Lower bounds on system zComputational requirements: ysum up process requirements over least-common multiple of periods, average over one period. z Communication requirements: yCount all transmissions in one period.

17. © 2000 Morgan Kaufman Overheads for Computers as Components Hardware platform design zNeed to choose: ynumber and types of PEs; ynumber and types of networks. zEvaluate a platform by allocating processes, scheduling processes and communication.

18. © 2000 Morgan Kaufman Overheads for Computers as Components I/O-intensive systems zStart with I/O devices, then consider computation: yinventory required devices; yidentify critical deadlines; ychooses devices that can share PEs; yanalyze communication times; ychoose PEs to go with devices.

19. © 2000 Morgan Kaufman Overheads for Computers as Components Computation-intensive systems zStart with shortest-deadline tasks: yPut shortest-deadline tasks on separate PEs. yCheck for interference on critical communications. yAllocate low-priority tasks to common PEs wherever possible. zBalance loads wherever possible.

20. © 2000 Morgan Kaufman Overheads for Computers as Components Internet-enabled embedded system zInternet-enabled embedded system: any embedded system that includes an Internet interface (e.g., refrigerator). zInternet appliance: embedded system designed for a particular Internet task (e.g. email).

21. © 2000 Morgan Kaufman Overheads for Computers as Components Examples zLaser printer. zPersonal digital assistant (PDA). zHome automation system.

22. © 2000 Morgan Kaufman Overheads for Computers as Components Example: Javacam zHardware platform: yparallel-port camera; yNational Semi NS486SXF; y1.5 Mbytes memory. zUses memory-efficient Java Nanokernel.

23. © 2000 Morgan Kaufman Overheads for Computers as Components Javacam architecture Web browser QuickCam 486 Java nanokernel Java VM HTTP Quickcam server QuickCam applet