Communication Overhead
There are n input ports within each IM and each port has at
most one cell arrival during a time slot, thus a VOMQ may
receive up to n arriving cells. Thus, a VOMQ should send a
dlogne-bit message to the central controller every time slot
for it to update the counter matrix
For our reference architecture, an IM has 144 VOMQs, and
each will generate an 8-bit message in each time slot. Given
that the cell size is 64 bytes, the traffic load of the control plane
is 1/64 of the data plane, where the port rate is about 10 Gbps.
There are totally 1442 VOMQs, so the I/O capacity of the
scheduler should be 1442 10=64 Gbps ¼ 3:17 Tbps. However,
if we assume that the I/O capacity of the scheduler is the
same as the VSC1433 chip, i.e., 1.4 Tbps, then the scheduler
will consist of at least three chips. If we further consider the
interconnections among them, the situation becomes more
complicated and seems infeasible to implement with the current
constraints. Thus, it is urgent to reduce the communication
overhead to make the scheduler feasible.
Communication OverheadThere are n input ports within each IM and each port has atmost one cell arrival during a time slot, thus a VOMQ mayreceive up to n arriving cells. Thus, a VOMQ should send adlogne-bit message to the central controller every time slotfor it to update the counter matrixFor our reference architecture, an IM has 144 VOMQs, andeach will generate an 8-bit message in each time slot. Giventhat the cell size is 64 bytes, the traffic load of the control planeis 1/64 of the data plane, where the port rate is about 10 Gbps.There are totally 1442 VOMQs, so the I/O capacity of thescheduler should be 1442 10=64 Gbps ¼ 3:17 Tbps. However,if we assume that the I/O capacity of the scheduler is thesame as the VSC1433 chip, i.e., 1.4 Tbps, then the schedulerwill consist of at least three chips. If we further consider theinterconnections among them, the situation becomes morecomplicated and seems infeasible to implement with the currentconstraints. Thus, it is urgent to reduce the communicationoverhead to make the scheduler feasible.
การแปล กรุณารอสักครู่..
