vrouter: Extend the flow hold count per-cpu statistics to 128 cpus
Affects | Status | Importance | Assigned to | Milestone | ||
---|---|---|---|---|---|---|
Juniper Openstack | Status tracked in Trunk | |||||
R1.1 |
Fix Committed
|
High
|
Anand H. Krishnan | |||
R2.0 |
Fix Committed
|
High
|
Anand H. Krishnan | |||
R2.1 |
Fix Committed
|
High
|
Anand H. Krishnan | |||
R2.20 |
Fix Committed
|
High
|
Anand H. Krishnan | |||
Trunk |
Fix Committed
|
High
|
Anand H. Krishnan |
Bug Description
vRouter keeps a per-cpu flow hold count statistic. This statistic is exported to user space processes to aid in debugging. While this statistic is maintained for all the cpus, vRouter exports statistics only for the first 64 cpus. The main reason why we limit the export to only 64 cpus is that the messaging infrastructure within vRouter has a notion of how much to allocate for each message based on the structure name. This calculation is not dynamic since for most structures the calculation remains the same and till now we have not felt the need. For flow, we allocate only for 64 cpus.
While making this calculation dynamic is a larger effort, for now we will extend the memory allocated to accommodate 128 cpus (which is reasonable). Also, split the regular flow request and the flow table information requests so that we allocate only what we need.
information type: | Proprietary → Public |
Review in progress for https:/ /review. opencontrail. org/10826
Submitter: Anand H. Krishnan (<email address hidden>)