Reference Guide

1. CSF — Output queues going from the CSF.
2. FP Uplink — Output queues going from the FP to the CSF IDP links.
3. Front-End Link — Output queues going from the FP to the front-end PHY.
All ports support eight queues — four for data traffic and four for control traffic. All eight queues are tunable.
Physical memory is organized into cells of 128 bytes. The cells are organized into two buffer pools — dedicated buffer
and dynamic buffer.
• Dedicated buffer — is reserved memory that cannot be used by other interfaces on the same ASIC or by other
queues on the same interface. This buffer is always allocated, and no dynamic re-carving takes place based on
changes in interface status. Dedicated buffers introduce a trade-off. They provide each interface with a
guaranteed minimum buffer to prevent an overused and congested interface from starving all other interfaces.
However, this minimum guarantee means the buffer manager does not reallocate the buffer to an adjacent
congested interface, which means that in some cases, memory is underused.
• Dynamic buffer — is shared memory that is allocated as needed, up to a configured limit. Using dynamic buffers
provides the benefit of statistical buffer sharing. An interface requests dynamic buffers when its dedicated
buffer pool is exhausted. The buffer manager grants the request based on three conditions:
– The number of used and available dynamic buffers.
– The maximum number of cells that an interface can occupy.
– The available packet pointers (2k per interface). Each packet is managed in the buffer using a unique
packet pointer. Thus, each interface can manage up to 2k packets.
You can configure dynamic buffers per port on both 1G and 10G FPs and per queue on CSFs. By default, the FP dynamic
buffer allocation is 10 times oversubscribed. For the 48-port 1G card:
• Dynamic Pool= Total Available Pool(16384 cells) — Total Dedicated Pool = 5904 cells
• Oversubscription ratio = 10
• Dynamic Cell Limit Per port = 59040/29 = 2036 cells
773