3750 Stack

[gagamboy15]

Hi Guys,

I just want to clarify something, does the 4x 3750 stack is faster than 6x 3750 stack? Does the the switch in 3750 stack share the same CPU, memory so it can perform faster?

Thanks in advance.

Cheers!

[Danilo Dy · Accepted solution]

Well, they share the 32Gbps switching stack interconnect. If you design your VLAN properly, i.e. one VLAN in one or two physical switch - don't put one VLAN to all the switches, unless otherwise you only have one VLAN.

Each switch will use its own CPU/RAM. The master switch serves as the control center for the stack and not central CPU/RAM for all member of the stack.

The master switch acts as the primary point of contact for IP functions such as Telnet sessions, pings, command-line interface (CLI), and routing information exchange.

The master is responsible for downloading forwarding tables to each of the subordinate switches. Multicast and unicast routing tasks are implemented from the master. QoS and access control list (ACL) configuration information is distributed from the master to the subordinates. When a new subordinate switch is added, or an existing switch removed, the master will issue a notification of this event and all the subordinate switches will update their tables accordingly.

[gagamboy15]

Thanks Medan.

So if the master is responsible for forwarding the tables, routing, access-list and qos,then there is a possibility that the master cannot handle all this features properly? It may impact LAN performance?

Thanks in advance.

[gnijs · Accepted solution]

No. All switching and forwarding is done locally on each individual switch and in hardware. The same for QOS.

What the previous poster wants to explain, is that the master switch runs the "management" protocols of the stack: it runs the telnet deamon, it runs the ospf routing process to determine the CEF tables. If routing changes, the master changes the forwarding tables of the clients, BUT forwarding on the clients is still done in hardware by the client switches itself (based on the tables).

The only performance problems you can have on a stack is: insufficient backplane capacity (32 gpbs, never experienced this) and -of course- bugs that prevent the master from working properly.

But simply adding a switch, will not "sufficate" the master or increase the load on the CPU master in anyway (other than the fact that it needs to update one extra switch)

[Joseph W. Doherty]

Geert's touched on an important point, possible insufficient backplane capacity, or perhaps to be more precise, insufficent ring capacity.

As I understand 3750 Stackwise, every switch in the stack places ALL its packets onto the ring. Further, it's the source member that also removes packets off the ring. In other words, each 3750 stack member floods the ring. More stack member, likely busier ring.

Also, the 3750 stack ring is marketed as 32 Gbps, but what it's really is dual duplex 8 Gbps links (4 x 8 = 32). Assuming that packets are placed on both rings, for redundancy, you also have cut your usable bandwidth in half.

So, in answer to your orignal question, a smaller stack might not be faster, per se, but there would be less stack memebers sharing the stack's bandwidth.

PS:

BTW:

The 3750-E besides doubling the ring's bandwidth (dual duplex 16 Gbps) with StackWise+, is much "smarter" in how the ring is used. First, only unicast packets that need to go off the stack member are placed on the ring. Second, the destination member strips the packet of the ring. These two improvements might support a much higher usage within Stackwise+ without even taking advanage of the extra physical bandwidth.