We're considering increasing the hold-queue size on an interface showing excessively high output drops from the default 40 to something like 100 or 200. Will this command have any effect on a 6509-VSS running 12.2(33)SXI3? We've exhausted all other options and tried to spread out our ASIC load as much as possible, but we think that sometimes congestion in other areas of the switch is contributing to microbursts.
When experiencing high output drops on 6509 without QoS enabled, will hold-queue xxx out be respected
cisco-catalyst
Best Answer
Without knowing much about your topology, I'll give you two things to consider:
1) Are the output queue drops noticeable to whatever the higher level application is? Often times we'll chase things in that we saw in our network management tool that look bad, but aren't really causing problems. TCP is designed to behave like this - push the network to the limit and then back off. So if you have a reasonable QoS policy in place to ensure that priority (generally real-time) packets aren't being dropped then output queue drops can just be par for the course.
2) If you are noticing issues related to the output queue drop then your options are limited. There is fundamentally an oversubscription problem here - likely you're feeding multiple 1G down a single 1G pipe or perhaps a 10G is down-stepping to a 1G pipe. IIRC the 6500 has something around 2Mb of buffer per port, and you're overrunning that continuously. There isn't a software command that's going to fix this for you (assuming you are already prioritizing the appropriate traffic down this link) - its time to increase the bandwidth. Again we don't know much about your environment, but look at moving from 1G to 10G, or to creating a port channel.