Hi,
I was tasked to find out how many PPPoE session an ASR9902 running 24.2.2 can handle. Oviously the scalability limits we can find on datasheets are one-dimensional and assume all the resources of the router are being employed towards that service. I get that its not possible to come up with a number that applies to every use-case out there.
The router is gonna run a mix of single stack PPPoE sessions (IPv6 will use PD) and dual stack sessions terminated on a PWHE. My current idea is to use ixia to test scalability limits for each single stack session then try to change the ratio of dual/single stack sessions to try and come up with some numbers tailored to our specific use-case.
I however, have no idea what should I be monitoring for on the router when it comes to check for resource usage, is it just TCAM entries?
Looking at pppoe logs during the creation of a single-stack ipv4 session I see:
LC/0/0/CPU0:Jun 13 10:36:53.813 CEST: pppoe_ea[341]: Need a total of 2 TCAM indexes
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Need 1 TCAM indexes for NP #0
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Allocated memory to hold 1 TCAM indexes for NP #0
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Need 1 TCAM indexes for NP #1
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Allocated memory to hold 1 TCAM indexes for NP #1
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Built list of 1 TCAM indexes for NP #0
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Built list of 1 TCAM indexes for NP #1
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Found tcam index 0x0 for index0 on np 0
LC/0/0/CPU0:Jun 13 10:36:53.814 CEST: pppoe_ea[341]: Found tcam index 0x0 for index0 on np 1
So now I know that a single IPv4 session is consuming 1 TCAM index for each NPU.
Looking at this doc:
https://www.cisco.com/c/en/us/td/docs/iosxr/asr9000/general-administration/configuration/b-general-administration-configuration-ios-xr-asr9000/scale-profiles-and-feature-profiles.html
I leanrn that these TCAM entries will be taken from the L2 table, partition 3 which by default has 1k entries. So my first question is how is the mapping between TCAM entries and indexes? My assumption was that 1 index = 1 entry but that would mean the router could handle a maximum of 1000 PPPoE sessions which is obviously too small of a number for an ISP. Even assuming the maximum 4k entries are all allocated for PPPoE single stack, that bumps up the number to 4k concurrent sessions which is still too small. The 9906 that are currently deployed as BNGs in our network are handling around 38k sessions.
I am assuming that TCAM limits will be hit way sooner than CPU/ram utilizations but the CLI provided to look at tcam usage:
show prm server tcam summary all all detail np0 location 0/0/CPU0
doesn't work on the 9902:
show prm server tcam summary all all detail np0 location 0/0/CPU0
Fri Jun 13 11:28:59.479 CEST
This command is not applicable to the node in the specified location
Is there a CLI we can use to gauge resource consumption by PPPoE sessions?
