"... In PAGE 10: ... Storage for vectors ~x and ~ y is allocated elsewhere. Table 1 and Table2 list the results of this benchmark for a LAN con guration consisting of two SPARCstation10 connected via Ethernet. Two processes exchange messages simultaneously with payloads varying from 4 KByte to 1 MByte.... In PAGE 10: ... The send daxpy counts are generated with the benchmark program given above, and the receive daxpy counts with the message continuation parametrized with the receive message (rmid). The di erence between Table 1 and Table2 is the application of marshaling: For the measure- ments presented in Table 1, an array of size payload is transferred without marshaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table 2 are generated with mar- shaling, where double arrays of size payload are marshaled by copying each double value into and out of the message.... In PAGE 10: ... The di erence between Table 1 and Table 2 is the application of marshaling: For the measure- ments presented in Table 1, an array of size payload is transferred without marshaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table2 are generated with mar- shaling, where double arrays of size payload are marshaled by copying each double value into and out of the message. This gives a notion of the overhead for marshaling dynamic data structures.... ..."

"... In PAGE 5: ... Storage for vectors ~ x and ~y is allocated elsewhere. Table1 and Table 2 list the results of this benchmark for a LAN configuration consisting of two Sun SPARC- station10 connected via Ethernet. The payloads of the si- multaneously exchanged messages varies from 4 KByte to 1 MByte.... In PAGE 5: ... The send-daxpy counts are generated with the benchmark program given above, and the receive- daxpy counts with the message continuation parameterized with the receive message (rmid). The difference between Table1 and Table 2 is the appli- cation of marshaling: For the measurements presented in Table 1, an array of size payload is transferred without mar- shaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table 2 are generated with marshaling.... In PAGE 5: ... The send-daxpy counts are generated with the benchmark program given above, and the receive- daxpy counts with the message continuation parameterized with the receive message (rmid). The difference between Table 1 and Table 2 is the appli- cation of marshaling: For the measurements presented in Table1 , an array of size payload is transferred without mar- shaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table 2 are generated with marshaling.... In PAGE 6: ... Furthermore, the band- width of the Ethernet is almost optimally utilized. For in- stance, according to Table1 , two 1 MByte messages are exchanged via the Ethernet within 1.849 seconds, which corresponds to a throughput of 9.... ..."

"... In PAGE 10: ... Storage for vectors ~x and ~ y is allocated elsewhere. Table1 and Table 2 list the results of this benchmark for a LAN con guration consisting of two SPARCstation10 connected via Ethernet. Two processes exchange messages simultaneously with payloads varying from 4 KByte to 1 MByte.... In PAGE 10: ... The send daxpy counts are generated with the benchmark program given above, and the receive daxpy counts with the message continuation parametrized with the receive message (rmid). The di erence between Table1 and Table 2 is the application of marshaling: For the measure- ments presented in Table 1, an array of size payload is transferred without marshaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table 2 are generated with mar- shaling, where double arrays of size payload are marshaled by copying each double value into and out of the message.... In PAGE 10: ... The send daxpy counts are generated with the benchmark program given above, and the receive daxpy counts with the message continuation parametrized with the receive message (rmid). The di erence between Table 1 and Table 2 is the application of marshaling: For the measure- ments presented in Table1 , an array of size payload is transferred without marshaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table 2 are generated with mar- shaling, where double arrays of size payload are marshaled by copying each double value into and out of the message.... In PAGE 11: ... Furthermore, the bandwidth of the Ethernet is almost optimally utilized. For instance, according to Table1 , two 1 MByte messages are exchanged over the Ethernet within 1.849 seconds, which corresponds to a throughput of 9.... ..."

"... In PAGE 52: ...n LE Client that conforms to this specification. The first is based on ISO 8802.3/CSMA-CD (IEEE 802.3) and has the format shown in Table7 below. The minimum LAN Emulation AAL5 SDU length for IEEE 802.... ..."

"... In PAGE 5: ...49 1690 2. 023 Table5 . Characterization of the communication performance on the two platforms.... ..."

"... In PAGE 5: ... Storage for vectors ~ x and ~y is allocated elsewhere. Table 1 and Table2 list the results of this benchmark for a LAN configuration consisting of two Sun SPARC- station10 connected via Ethernet. The payloads of the si- multaneously exchanged messages varies from 4 KByte to 1 MByte.... In PAGE 5: ... The send-daxpy counts are generated with the benchmark program given above, and the receive- daxpy counts with the message continuation parameterized with the receive message (rmid). The difference between Table 1 and Table2 is the appli- cation of marshaling: For the measurements presented in Table 1, an array of size payload is transferred without mar- shaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table 2 are generated with marshaling.... In PAGE 5: ... The difference between Table 1 and Table 2 is the appli- cation of marshaling: For the measurements presented in Table 1, an array of size payload is transferred without mar- shaling (marshal = unmarshal = (void *) NULL). In contrast, the values shown in Table2 are generated with marshaling. Here arrays of double precision numbers of size payload are marshaled by copying each double preci- sion value into and out of the message.... ..."