09.07.2020

Cisco Switches

I would like to remind you that in the old ASIC’s (in the 3750E / 3750X switches) the elements responsible for processing incoming packets and the elements responsible for processing outgoing packets are separated. They do not have a common bus inside the ASIC. Therefore, even if a packet is transmitted between ports served by the same ASIC, this packet will necessarily be sent to the switching factory.

The second difference with the StackWise-160 is its increased stack bus bandwidth. The bandwidth of the stack cable is now 40 Gb / s (full duplex). Thus, the bandwidth of the entire stack bus for StackWise-160 technology: 40 Gbit / s * 2 (in each direction) * 2 (number of ports) = 160 Gbit / s It should be noted that, unlike the 3750 series of switches, the stack kit for 3650 sold separately. For the StackWise-160 technology, the general scheme of the stack at the software level has been changed. Stateful Switch Over (SSO) is now used to provide fault tolerance. As we recall, previous technologies (StackWise and StackWise Plus) use a simpler failover scheme. One of the switches is selected as the stack master. It performs logical operations (control-plane) for the entire stack. Only hardware tables (MAC tables and CEF tables (FIB / Adj)) are synchronized between the switches on the stack. The remaining tables, including the routing table, are re-filled on the new wizard. Those. control-plane starts from scratch. At 3650 switches, a more advanced scheme, Nonstop Forwarding with Stateful Switchover (NSF / SSO), has been used to provide fault tolerance. There is no longer such a thing as a master. Now the Active-Standby scheme is used. One of the switches is selected by the main (Active), another one by its hot standby, which synchronizes with the main all the necessary information (L2 and L3). Control-plane now works in Active-Standby mode. This minimized the time required for recovery in the event of a failure of the main switch.

Stackwise-480

Let’s take a look at the StackWise-480 technology. With it, you can stack 3850 series switches.

The 3650 and 3850 switches are very similar. Both of these series are based on UADP ASIC. Accordingly, the algorithms of the StackWise-480 and StackWise-160 stack are similar. True, there is a difference. StackWise-480 uses three physical stack rings. This is achieved by the fact that inside one stack cable for 3850 switches, there are three wires (Fig. 6). Each with a bandwidth of 40 Gb / s (full duplex). The bandwidth of the entire stack bus for StackWise-480 technology:

40 Gbps * 2 (each way) * 3 (number of wires) * 2 (number of ports) = 480 Gbps

At the logical level, the stack is represented by six paths (two logical paths per wire). Packets along three logical paths “spin” in one direction, and along three other paths – in the other (Fig. 7).

The choice of the path is carried out in the same way as before, using tokens.

Flexstack

With this review of the Stackwise family of stacking technologies, I propose to end it. Let’s take a look at the 2960 family of switches and the FlexStack and FlexStack Plus stacking technologies.

Stacking for 2960 first appeared on 2960-S switches. To combine switches using FlexStack technology, a stack module and specialized cables with a bandwidth of 10 Gb / s (full duplex) are used. Each stack module has two ports. When combined, the switches are connected in a ring (although this is not necessary). The total throughput of the entire stack bus is:

10 Gbit / s * 2 (in each direction) * 2 (number of ports) = 40 Gbit / s.

For the FlexStack stack, packet-to-packet transfer between switches occurs device-by-device. The switch for each packet determines where to send it: to the regular or to the stack ports. This interaction resembles the operation of several switches connected to each other via Ethernet. The difference is that the communication between the stack switches is provided by the FlexStack protocol. The choice of which stack port to send the packet through is determined by a special algorithm that resembles the operation of OSPF. Those. the shortest path to the switch in the stack on which the destination port is located is selected. If any changes occur (for example, one of the switches failed or the stack cable is disconnected), this algorithm recounts the paths again.

Leave a Reply

Your email address will not be published. Required fields are marked *