Packet buffer management

The following table lists the packet memory bandwidth and the total packet memory on ICX devices.

Table 1. Packet memory on ICX devices
ICX device Total bandwidth Total packet memory
Ruckus ICX 7150 126 Gbps 2 MB
Ruckus ICX 7250 200 Gbps 4 MB
Ruckus ICX 7450 200 Gbps 4 MB
Ruckus ICX 7650 480 Gbps

564 Gbps (IO 48ZP/48ZF)

328 Gbps (IO 48P)

8 MB
Ruckus ICX 7850 2.0 Tb/s

3.2 Tb/s

32 MB

These devices run in cut-through mode, which means that cut-through eligible packets are not buffered. If a packet must be buffered, it is buffered after Layer 2 and Layer 3 lookup. The packet priority is classified before buffering.

Note: The ICX 7150, ICX 7250, ICX 7450, and ICX 7650 devices do not run in cut-through mode.

There are two independent packet admission mechanisms: ingress buffer management and egress buffer management.

Ingress buffer management

On the Ruckus ICX 7150, ingress buffer management tracks buffer utilization on a per-ingress-port basis.

  • Loss-less behavior through symmetric flow control is supported.
  • Buffers are reserved for high-priority traffic.

As these accounting structures reach their limit, incoming packets to the ingress port are dropped.

On the Ruckus ICX 7250, ICX 7450, and ICX 7850 devices, ingress buffer management determines whether a packet should be admitted into memory based on the state of available memory and the amount of buffer resources in use by the ingress PG. The aim of the mechanism is to support fair access to buffering resources while also enabling loss-less operation across a network. The memory is logically divided into three sections:
  • Guaranteed
  • Shared
  • Headroom for flow control in on-the-fly packets

Ingress buffer limits are automatically configured based on your configuration to support either loss-less or tail-drop operation. You can configure the qos ingress-buffer-profile command to configure a share level, which determines the maximum number of buffers a PG can use as a fraction of the total sharing pool. For example, if PG 0 is at level 4, it can use up to 1/9 of the total sharing buffers in the sharing pool. The actual number of buffers a PG can use depends on the number currently available in the system.

On the Ruckus ICX 7650 and the Ruckus ICX 7850, there is a default profile for ingress buffer management, but it is not configurable because PFC is not supported.

Egress buffer management

This mechanism tracks buffer utilization on a per-egress port and priority basis. As these accounting structures reach the limit, packets that are destined to the congested egress port-priority are tail-dropped. The aim of the mechanism is to support fair access to the buffering resources among congested egress ports. Any incoming packet is counted only once per egress port regardless of whether it is unicast or multicast. Memory is logically divided into two sections:
  • Guaranteed is on a per-port-priority basis.
  • Shared is on a per-port basis for the Ruckus ICX 7150 device. It is on a per-port-priority basis for the Ruckus ICX 7250, ICX 7450 and ICX 7850 devices.

On the Ruckus ICX 7250, ICX 7450, ICX 7650 and ICX 7850 devices, sharing is a ratio of the remaining buffers. You can configure the share level to determine the maximum number of buffers that an egress queue can use as a fraction of the total sharing pool. For example, if queue 4 is at level 4, it can use up to 1/9 of the total sharing buffers in the sharing pool. You can configure eight levels of sharing. The actual number of buffers that a queue can use depends on the number currently available in the system.

On the Ruckus ICX 7150, buffer sharing is on a per port basis instead of a per queue basis. Also, the buffer sharing level is calculated with a fraction of the total number of buffers. For example, level7-1/2 allows 1/2 of the total buffers as sharing buffers, which is 1 MB.

The ICX 7150, ICX 7250, ICX 7450, and the ICX 7650 devices run in store-and-forward as the default mode for packet forwarding. In the ICX 7850 device, 1 Gig always operates in store-and-forward mode, irrespective of the global switch mode configuration.