FBFM2112F897CSLJLS View Datasheet(PDF) - Intel

Part Name

Description

Manufacturer

FBFM2112F897CSLJLS

Intel® FM2112 24-Port 10G/1G Ethernet Switch Chip

Intel 

Intel® FM2112 24-Port 10G/1G Ethernet Switch Chip Data Sheet

3.3.2.1

3.3.3

• The shared pool status specifies how full the memory is that is shared between the

different ports. The total available shared pool is defined as the total memory

minus the sum of each ports private memory.

The Intel®Ethernet Switch Family supports the following watermarks,

• RX-Private (per port, both Pause and PWD)

— Frames from the ith RX port may use ith RX-Private queue.

— The sum of the RXi-Private total memory (1MB).

— RX-Private is the same for both Pause and PWD

• RX-Shared and TX-Shared watermarks for Pause and PWD

— Shared watermarks are “Hog watermarks” and once the occupancy exceeds the

watermarks, either the ports are paused or the frames are dropped with 100%

probability. Note that the pause or drop decision for a frame is made based

upon the queue occupancy at its time of arrival and before that frame is added

to that queue. If a WM is set to 32, for example, the 33d frame will not be

paused/dropped because the WM has not been exceeded. The 34th frame to be

considered for that queue will be dropped/paused because the WM has now

been exceeded.

— While an RX queue occupancy is between RX-Private and RX-Shared, the switch

may pause the RX port or drop frames for PWD.

— The user must set RXi-Shared > RXi-Private.

• Global PWD watermarks

— Low - The lower PWD watermark.

— High - The upper PWD watermark.

• Global Privileged watermark

— Prevents MAC overflow

Queue configuration

For all of the watermarks, the queue size is an integral number of 1024

bytes. The size 1024 bytes is a convenience of the PWD and Pause

processing, and does not reflect the segment size of the memory.

PWD (Priority Weighted Discard)

The FM2112 uses a PWD (Priority Weighted Discard) to protect queue

resources preferentially for higher priority tagged frames. Figure 11

shows a queue without PWD and a queue with the FM2112's

implementation of PWD. The solid/dotted lines represent different PWD

priorities, where the different priorities begin 100% drop at user

assigned queue occupancy levels. This makes the PWD implementation

a superset of the simple queue.

46

Share Link: