Barton, E., Cownie, J., and McLaren, M. (1994). Message Passing on the Meiko CS-2. Parallel Computing, 20(4).  Home/Search   Document Not in Database   Summary   Related Articles   Check

....this approach has the advantage of not requiring any hardware support, so it can be implemented on any machine. Another approach is to tag messages with a job ID. When an arriving message does not belong to the currently running job, it is handled anyway. This approach is used on the Meiko CS 2 [47] and the Intel Paragon [461] Note that it requires access to the descheduled job s address space, and that the network adapter (or communication coprocessor) must be smart enough to generate and check the job IDs. Otherwise, user level communication cannot be used. Security issues are resolved by ....

E. Barton, J. Cownie, and M. McLaren, "Message passing on the Meiko CS-2 ". Parallel Comput. 20(4), pp. 497--507, Apr 1994.

....this approach has the advantage of not requiring any hardware support, so it can be implemented on any machine. Another approach is to tag messages with a job ID. When an arriving message does not belong to the currently running job, it is handled anyway. This approach is used on the Meiko CS 2 [30] and the Intel Paragon [281] Note that it requires access to the descheduled job s address space, and that the network adapter (or communication coprocessor) must be smart enough to generate and check the job IDs. Otherwise, user level communication cannot be used. Security issues are resolved by ....

E. Barton, J. Cownie, and M. McLaren, "Message passing on the Meiko CS-2 ". Parallel Comput. 20(4), pp. 497--507, Apr 1994.

....incoming messages away in a separate queue and brings them to job i s attention as soon as it is re scheduled to run. The hardware may even place the messages in the appropriate spot in user memory. Such a hardware mechanism may be seen (in the so called network interface chip) in Meiko s CS 2 [3] and Intel s Paragon [24] Gang Scheduling: When processors switch context (all in unison) the communication state is saved by draining the router network all messages are sent quickly to some PE or the other, even though they may reach wrong destinations. 16 This leaves the next job with ....

E. Barton, J. Cownie, and M. McLaren. "Message Passing on the Meiko CS-2". Parallel Computing, vol 20(no 4):497--507, April 1994.

....based MPP systems. These systems are typical message systems, and the most recent product, the CS 2, is among the most powerful MPP systems developed todate. 23 CS communication, outlined in Table 7, differs dramatically from other systems in that it is based on a communications name space [8]. Processes create virtual communication objects called transports. Transports become useful only when registered with a global name server implemented by the function csnregname. Once a transport has a registered name, other processors that know of the name can look up the name and, if ....

E. Barton, J. Cownie, M. McLaren, "Message Passing on the Meiko CS-2", later in this proceedings.

....overhead 80386 100 byte message (18 from context switches) 110 S[19] with MPC 100 byte message (Communication Co Processor) CM 5 32 Mhz SPARC 143 S swap[9] CMMD library 3. 4 S swap[9] CMNF library J Machine 12.5 Mhz MDP 43 cyc[10] Streaming Injection 1024 max round trip null RPC CS 2 40 Mhz 20 S[39] Channel SPARC 24.6 S[23] DMA w active message Hardware Table Lookup 174 S[21] PARMACS macros ping pong 206 S[20] mpsc library mesg exhange T3D 150 Mhz 21064 600nS[26] Shared Memory 2048 max remote read 2:76 S [40] Fast Messages F I Specific 16 byte Fetch and Increment Hardware Support ....

Eric Barton, James Cownie, Moray McLaren, "Message Passing on the Meiko CS-2", in Parallel Computing 1994, pp. 97--507.

No context found.

Barton, E., Cownie, J., and McLaren, M. (1994). Message Passing on the Meiko CS-2. Parallel Computing, 20(4).

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC