06-01-2011 07:20 AM
Does Blade's Switch can have port oversubscription issue (it has 20 ports) ?
As well, what is the command to see the restrictions (like "show port-resources module #" on MDS)
06-01-2011 02:43 PM
a 4020 blade switch naturally has oversubscription as it bridges 14 internal ports to 6 external ones. Depending on application workloads you might want to achieve oversubscription rates somewhere between 7:1 and 3:1.
Unless some throughput requirements are already defined, my best practice is to start with two external links (for redundancy rather than performance) and add more if necessary.
The best indication of link congestion is tim_txcrd_z counter found in portShow output. When this counter grows rapidly, this means that port can't keep up sending traffic (because most likely the device on the other side can't keep up receiving and processing it).
Hope this helps,
06-02-2011 01:52 AM
as I saw, tim_txcrd_z is an indication of Time of BB credit zero.
Can it indicate on another performance issues?
Is there another way to see current subscription status ?
06-07-2011 10:09 AM
tim_txcrd_z indicates a bottleneck in the environment. This essentially means that something is not capable enough to receive the number of frames being sent to it. Can be an oversubscription problem on a given segment as well as a slow endpoint device.
Theoretical oversubscription is calculated as downstream (clients) to upstream (network) bandwidth ratio. This may or may not reflect actual oversubscription depending on nodes activity and traffic patterns.
06-08-2011 01:33 PM
Nope. Cisco had that feature because their first SAN switches were oversubscribed on port group level and they had to deal with that somehow.
You can monitor port load in realtime with SNMP or DCFM.