Saturday, February 4, 2012

Cisco 3750 2960 Switch Stacking Reference


Figure 1. FlexStack Comparison


Figure 2. FlexStack with Redundant Links



Figure 3. FlexStack Module



Deployment Topologies

Figure 5 shows a three-member stack with full bandwidth and with redundant FlexStack connections. Each member in Figure 5 has 20Gbps of stack bandwidth and is fully redundant.

Figure 5. Fully Redundant Three-Member Stack


When one FlexStack Link is not present, the stack still functions. Figure 6 shows a stack with incomplete FlexStack cabling. In Figure 6, all data traffic passes through the middle member. This stack is operating in a nonredundant mode. This stack provides only half of the possible bandwidth between members and does not have redundant connections. Only the middle member has 20Gbps stack bandwidth. The top and bottom members are operating at 10Gbps stack bandwidth.

Figure 6. Nonredundant Three-Member Stack


Three different FlexStack cable lengths are available, shown in Table 1.

Table 1. Cable Lengths


Product ID

Length

Description

CAB-STK-E-0.5M

0.5 meters

This is the default cable that ships with the FlexStack module.

CAB-STK-E-1M

1.0 meters

CAB-STK-E-3M

3.0 meters

The different lengths allow for deployment flexibility. Figure 7 shows how to stack four switches using only the 0.5-meter cables. The stack member connections are interweaved using the stack cables. No cable extends more than two stack members. The interweaving of the stack links still provides redundant connections for the stack.

Figure 7. Four Member Stack with 0.5-Meter Cables


Figure 8 shows the conventional cabling deployment. The 3.0-meter cable is used to complete the redundant ring connection by connecting the top member with the bottom member. The other connections connect to directly adjacent members using the 0.5-meter cables. The 3.0-meter cable is not the default cable shipped for FlexStack and must be ordered. See the Cisco Catalyst 2960-S data sheet for ordering information.

Figure 8. Four Member Stack with 3.0-Meter Cable


Figure 9. Adding Members to an Existing Stack




Figure 9. Recommendation of Four member stack with 2 Uplinks


http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps6406/white_paper_c11-578928.html

Table 2. Comparing FlexStack and StackWise Plus
 
Cisco Catalyst 2960-S Stacking Ease of Use
Cisco Catalyst 3750-X StackWise Plus
Ease of Use and High Availability
Device limit
4 units
9 units
Stack bandwidth
10G/20G
32G/64G
Architecture
Hardware drop table
Ring (destination stripping)
Dynamic ring load balancing
No
Yes
Stack convergence
1-2 seconds
Few milliseconds
Stack quality of service (QoS)
Applied hop by hop
Applied on ingress
Management
Single IP address, SNMP, syslog
Single IP address, SNMP, syslog
Configuration
Single config and CLI, auto image and config update
Single config and CLI, auto image and config update
Show and debug commands
Unified
Unified
Single forwarding and control plane
Synchronize Address Resolution Protocol (ARP), MAC address, Internet Group Management Protocol (IGMP), VLAN tables
Synchronize Address Resolution Protocol (ARP), MAC address, Internet Group Management Protocol (IGMP), VLAN tables, routing tables
Cross-stack features
Yes
Yes
Single bridge-ID
Yes
Yes
Redundancy
Stack master 1:N redundancy
Stack master 1:N redundancy
The differences between FlexStack and StackWise Plus are rooted in the different hardware architectures. StackWise Plus is the superior architecture. StackWise Plus is a dual redundant ring architecture with individual links operating at 32Gbps. The StackWise Plus connections are not pure Ethernet links but rather a ring architecture, allowing packets to take either ring. The FlexStack links are full duplex 10Gbps Ethernet links. The ring architecture gives StackWise Plus a significant advantage over FlexStack when it comes to high availability of the stack.
In Table 2 stack convergence is measured in milliseconds for StackWise Plus, as opposed to seconds for FlexStack. This is because of the ring architecture. In StackWise Plus packets ingressing the stack can be forwarded on either ring. When one ring becomes inoperable, all ingressing packets are forced to a single ring. The speed of one ring is 32Gbps, and can easily accommodate an entire stack. The recovery time for a single link going inoperable is not dependent upon the number of members in the stack. It is consistent across all stack sizes. In FlexStack, when a single link becomes inoperable, there is a stackwide disruption in the forwarding of packets. The members must rediscover the new topology. This rediscovery time can take up to 1-2 seconds depending upon the number of members in the stack. Smaller stacks will reconverge faster than larger stacks.
Because FlexStack stack links are Ethernet links, forwarding packets from one stack member to another requires the packets to be forwarded by each member. This is referred to as hop by hop, meaning the packets must hop from the ingressing member to the egressing member across all members in between. Not all members see the packet. Only those members in the data path will forward the packet. This hop-by-hop behavior is the same as different Ethernet switches forwarding the packet amongst themselves (at 10Gbps speeds). Hop by hop requires that each member must queue the packet. Queuing the packet means that QoS is applied at each hop. If a stack link between two members becomes congested, packets that successfully ingress on one member may be discarded by another member. There is no dynamic adjustment to the forwarding of packets when one link becomes congested. With StackWise Plus, once a packet is on the ring, it is guaranteed to be delivered to the destination interface. Add to this that StackWise Plus rings will dynamically load balance, and you can see that ring congestion is not an issue on StackWise Plus.


Cisco 3750 Access Switch Stacking Quick Reference

Operational Notes:
The stack master is the single point of stack-wide management. From the stack master, you configure:
- System-level (global) features that apply to all stack members
- Interface-level features for each stack member
All stack members are eligible stack masters. If the stack master becomes unavailable, the remaining stack members participate in electing a new stack master from among themselves. A set of factors determine which switch is elected the stack master. One of the factors is the stack member priority value. The switch with the highest priority value becomes the stack master.
A higher priority value for a stack member increases its likelihood to be elected stack master and to retain its stack member number. The priority value can be 1 to 15. The default priority value is 1.
The stack master contains the saved and running configuration files for the switch stack. The configuration files include the system-level settings for the switch stack and the interface-level settings for each stack member. Each stack member has a current copy of these files for back-up purposes.
You manage the switch stack through a single IP address. The IP address is a system-level setting and is not specific to the stack master or to any other stack member. You can manage the stack through the same IP address even if you remove the stack master or any other stack member from the stack.
A switch stack has up to nine stack members connected through their StackWise ports. A switch stack always has one stack master.
Command Usage Notes:
You can display the stack member number by using the user EXEC command:
Switch#show switch
You can manually change the stack member number by using the global configuration command:
Switch(Config)#switch (current-stack-member-number) renumber (new-stack-member-number)
The new number goes into effect after that stack member resets.
A single switch can be reset in the stack by using the privileged EXEC command:
Switch#reload slot (stack-member-number)
You can change the priority value for a stack member by using the global configuration command:
Switch(Config)#switch (stack-member-number) priority (new-priority-value)
To manually upgrade a stack member using the IOS from another stack member:
Switch#archive copy-sw /destination-system (destination-stack-member-number)
/force-reload (source-stack-member-number)
To run a command (such as show version) on a single switch in the stack:
remote command (stack-member-number) show version
Valid values for stack-member-number are 1-9

No comments: