8. Performance > Port-to-Port Performance Characteristics
195
Tsi578 User Manual
June 6, 2016
Integrated Device Technology
www.idt.com
8.4.2.2
Many Ports-to-One Port Throughput Performance
Under a non-congested, many ports-to-one port packet traffic scenario, when all of the total ingress line
rates are the same as the egress line rate (for example, four 1x mode, 3.125 Gbaud ingress ports all
going to one 4x mode, 3.125 Gbaud egress port), the ingress port and egress port will always maintain
line rates. This means there will be no retry of packets at the ingress and no bubble occurring in the
egress packet streams except for the idle sequence insertion every 5000 code-groups required by the
RapidIO Interconnect Specification (Revision 1.3)
. This is true for any payload size and different
priorities. The arbitration scheme within the device allocates sufficient bandwidth for each ingress port.
When the total of the ingress line rates exceed that of egress port, retries occur at one or more of the
ingress ports if the packet density exceeds the capacity of the egress port. The egress port still
maintains its maximum packet rate with no bubble. This is true for any payload size and priorities.
8.4.2.3
One Port-to-Many Port Throughput Performance
Under a non-congested one port-to-many ports packet traffic scenario, when the ingress line rate is the
same as the total egress line rates (for example one 4x mode, 3.125 Gbaud ingress port splitting to four
1x mode, 3.125 Gbaud egress port), the ingress and egress always maintain line rates. This means there
is no retry of packets in ingress and no bubble-packet in the egress packet streams except for the idle
sequence insertion every 5000 code-groups required by the RapidIO specification. This is true for any
payload size and different priorities. The arbitration scheme within the device divides the traffic
according to the egress port bandwidths.
When the ingress line rate exceeds that of the total of the egress ports, retries occur at the ingress port
when the packet density exceeds the buffer capacity. The egress ports still maintains their maximum
packet rates with no bubble. This is true for any payload size and priorities.
8.4.3
Multicast Performance
8.4.3.1
Multicast Latency
Since multicast involves more than one egress port and each egress port can have independent traffic
conditions, a multicast packet can appear at the destination egress ports at different times. A minimum
multicast latency is defined as the shortest time from the arrival of the first bit of a packet at an ingress
port that will be multicast, to the appearance of the first bit of the multicast packet at an egress port
under no resource contention.