Results 1 - 10
of
21,727
Table 5: Node with lowest communication/synchronization time
TABLE I The synchronization algorithm for each node i, i negationslash = 1.
Table 7: Summary of detailed information of nodes with maximum communication/synchronization time on several platforms (% of inclusive time)
Table 2: The synchronization server performance with varying number of participating nodes
Table 2 gives the timing for synchronization of nodes with FEMPI using the barrier synchronization function MPI_Barrier defined by the MPI standard. From the table, it can be seen that the synchronization time is low in FEMPI on the Xeon cluster. As expected, the synchronization time is higher on the embedded system, which can be attributed to the slow processing power and network.
"... In PAGE 6: ...5 8 128 2048 32768 Payload (Data size in KB) T h r o u g h p u t ( M b p s ) Xeon, Point-to-point (Send, Recv) Xeon, Point-to-point (Send, Recv), MPICH Xeon, Point-to-point (Sendrecv) Embedded, Point-to-point (Send, Recv) Embedded, Point-to-point (Sendrecv) Figure 2. Performance of point-to-point communications Table2 . Barrier synchronization in FEMPI No.... ..."
Cited by 1
TABLE I The synchronization algorithm for each node a227 , a227a102a228 a229a178a230 .
Table 1: Synchronization Table
"... In PAGE 18: ... However, putting S6 and S9 together relies on the fact that the false anti-dependence can be clearly disambiguated at compiler time. Figure 7 lists the corresponding stranded code and Table1 lists synchronization signals among strands. Please notice that merge and region-end nodes are eliminated since their existence is to help compute synchronization signals among threads.... In PAGE 19: ...i=0; SYNC(ST_1); END_STRAND(); STRAND_1: s0_1: CondSync((i lt;BuferSize),2, 9) END_STRAND(); STRAND_2: s1: symbol=Buffer[addr]; s2: w = Weight[i]; s3: r = 0; s4: list = list_header; s10: r1 = 0; s11: list1 = list_header1; END_STRAND(); STRAND_3: s5: SYNC(ST_5); s12: SYNC(ST_S12); /* Not Shown */ END_STRAND(); STRAND_5: s5_1: CondSync((list!=NULL),6, 4); END_STRAND(); STRAND_6: s6: entry=list- gt;symbol; s9: list = list- gt;next; END_STRAND(); STRAND_7: s7: CondSync((entry==symbol),8,5); END_STRAND(); STRAND_8: s8: r = r + w ; END_STRAND(); STRAND_4: s_17 Weight[i] = r + r1 ; s0_2: i = i + 1; END_STRAND(); STRAND_9: Figure 7: Threaded C code with explicit synchronization heuristics can be used to leave CondSync at the end of a strand so that the switching of strand must happen after this operation is issued. The rst two columns of Table1 is actually coded in the stranded executable code, and they are used to initialize the strand synchronization. The ring of a strand is managed by hardware.... ..."
Table 3. Comparison between context-based synchronization mechanism and non context-based synchronization mechanism
"... In PAGE 8: ...shown in Table3 . However, in case of the heterogeneous distributed systems consisting of N nodes, the context-based synchronization may be better than the non context-based.... ..."
Cited by 2
Table 1. Summary of the Synchronous Bandwidth Allocation Schemes.
1995
"... In PAGE 21: ... Table1 shows the worst case achievable utilizations of the synchronous band- width allocation schemes introduced above. These results were derived using the assumption that D i = T i , i.... In PAGE 21: ...y these allocation schemes. The remaining 0.67 of the usable network bandwidth can be used for the transmission of asynchronous messages. From Table1 , it can be seen that the normalized proportional scheme and the local scheme both have the same worst case achievable utilization. An advantage of the local allocation scheme, however, is that it only uses information local to node i in calculating the synchronous bandwidth H i .... In PAGE 22: ... Robustness.Asshown in Table1 , the worst case achievable utilization of many synchronous bandwidth allocation schemes is known. The characteristics of syn- chronous message schemes can be altered and all messages will still meet their deadlines, provided that the utilization of the synchronous message streams remains below the worst case achievable utilization.... ..."
Cited by 6
Results 1 - 10
of
21,727