IDT Switch Description
Revision 1.5
Integrated Device Technology, Inc.
CPS-16/12/8 User Manual
3 - 3
July 10, 2012
3.3.3 Switch Core
The switch core acts like a three stage switch composed of TDM, Mesh and TDM. Full mesh PVC connects
QUAD to QUAD, and QUAD to maintain block as well, TDM connect Ports to PVCs.
3.3.3.1 QUAD to QUAD Full Mesh PVC
The PVC module is used to connect every quad to every other quad as well as all quads to the mainte-
nance handler and the maintenance handler to all quads. It handles data transfer as well as control. The
PVC module serves as a pipeline connection between nodes. It provides no other functionality beyond this
and simply made up of registers for all incoming signals that then drive the outputs. It has all the advan-
tages of full mesh.
3.3.3.2 PORT to PVC TDM
Inside each QUAD, the PORTs are connected to PVC network in TDM manner. It is this time sharing
concept that is the origin of many of the sub-modules that refer to time division multiplexing (TDM) in regard
to PVC operation. This TDM method is strictly per PVC and is not functional as an overall switch-wide time
division scheme.
3.3.4 Output Buffers
There are separate output buffer resources for maintenance packet and data packet. The output buffer
provides a temporary storage for outgoing packets, it decouples the switching and transmitting. It achieves
wire speed while transmitting different priority packets per sRIO specification.
For data packet path, the output buffer size is 448 bytes per priority per port. Each output buffer can track
up to 3 packets, given that there is enough buffer space for them. The output buffer will only allow new
packet in if it has free tracking resources and buffer space available for a full maximum sized packet.
For maintenance packet path, the buffer size is 88 bytes per priority per port. The separate maintenance
packet buffer forms an independent data path.
3.3.5 Retransmit Buffers
There is 4-max-packet-size retransmit buffer for each priority for a given port. For a given port, there is
totally 16 max-packet-size buffer. Each priority can keep track of up to 32 packets. Both data packet and
maintenance packet share the same buffer. The retransmit buffer is enough to deal with normal response
delay.
3.4 SWITCHING SCHEDULER AND PRIORITIES
3.4.1 Input Buffer to Output Buffer
First arrive first served is the basic rule of moving data from input buffer to output buffer. For the same
source port, strict-priority is applied. That means packets in the higher priority queue always is served first.
For a given queue of given source port, it can bypass the destination-blocked packet to serve subsequent
packets. Maintenance packets are always treated as higher priority than data packet.
For the same destination, same priority, different sources to this destination are done in a Round-Robin
manner.
These rules apply to multicast and unicast. The difference is that the input buffer will not be released until
the packet has been forwarded to all destinations in the multicast list. In another words, one blocked desti-
nation port of the multicast list will not block the forwarding to other destination ports (multicast splitting), but
it does hold the input buffer resource.
