On Thu, 25 Apr 2019 at 00:19, Martin T <m4rtn...@gmail.com> wrote: > in PFE) could indeed be affected. On the other hand, the overall LC > CPU usage according to "sh linux cpu usage" did not exceed 80% even > during the route churn, but I actually do not know what exactly this > utilization means..
Like ethernet, CPU core is either busy (100%) or not busy (0%), and load is looking some specific time frame and telling how often in that time frame CPU was busy. So 80% of time, work needed to compete for access to the CPU. If I understood your explanation right (I may not have): WireI => PFEi => RE => PFEe => WireE You measure stable delay from WireI to PFEe even under load (so PFEi is not contributing to jitter, neither is RE (running RPD) contributing on either receive or send direction). But you measure variant delay from WireI to WireE (so PFEe is largely responsible of jitter) So perhaps ingress side punt is done through LC CPU interrupt and egress side is not, causing egress side to cause more jitter, depending on what the egress PFE LC CPU is doing. -- ++ytti _______________________________________________ juniper-nsp mailing list juniper-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/juniper-nsp