Cisco Support Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Community Member

6148 and oversubscription

When would over subscription actually become an issue that I can show management via a show command on the switch.

When I look at the cards I see only output drops on the occupied ports. Utilization is low and there are no other errors yet users still complain the net is slow.

1 ACCEPTED SOLUTION

Accepted Solutions
Hall of Fame Super Silver

Re: 6148 and oversubscription

Hello Mark,

I would say that the output drops on the port are already meaningful because:

each output drop frame is simply lost, the IP packet carried inside the ethernet frame can be part of a TCP conversation.

So a dropped packet is detected at the TCP level and this requires retransmission.

when the probability of drop is not so small all user TCP sessions are affected and users complain of resulting slowness = low TCP goodput

This is the kind of reasoning you should offer to the management, if users complain of slowness it is a problem this is the point regardless that there is no red alarm on the module or other direct indicator.

I've found a useful link that can help you in explaining the issue when servers are involved (it should be in your case if I remember correctly)

http://www.cisco.com/en/US/products/hw/switches/ps700/products_tech_note09186a00801751d7.shtml#ASIC

it provides a show command

show counters           interface gigabitEthernet | include           qos3Outlost

and a workaround to reduce the effects of oversubscription (to be tested  so I would say that it can be not effective I wouldn't mention this to management)

Hope to help

Giuseppe

1 REPLY
Hall of Fame Super Silver

Re: 6148 and oversubscription

Hello Mark,

I would say that the output drops on the port are already meaningful because:

each output drop frame is simply lost, the IP packet carried inside the ethernet frame can be part of a TCP conversation.

So a dropped packet is detected at the TCP level and this requires retransmission.

when the probability of drop is not so small all user TCP sessions are affected and users complain of resulting slowness = low TCP goodput

This is the kind of reasoning you should offer to the management, if users complain of slowness it is a problem this is the point regardless that there is no red alarm on the module or other direct indicator.

I've found a useful link that can help you in explaining the issue when servers are involved (it should be in your case if I remember correctly)

http://www.cisco.com/en/US/products/hw/switches/ps700/products_tech_note09186a00801751d7.shtml#ASIC

it provides a show command

show counters           interface gigabitEthernet | include           qos3Outlost

and a workaround to reduce the effects of oversubscription (to be tested  so I would say that it can be not effective I wouldn't mention this to management)

Hope to help

Giuseppe

1536
Views
0
Helpful
1
Replies
CreatePlease to create content