On Wed, Oct 14, 2015 at 5:25 AM, Jamal Hadi Salim <j...@mojatatu.com> wrote: > On 10/12/15 14:38, Cong Wang wrote: >> >> It is odd to see qlen!=0 but backlog==0, for a real example: >> > > Backlog is a transient stat so a lot of times it should be 0. Only when > the CPU is sending faster than the link can handle should you see > the backlog grow (and eventually drain to 0).
Of course. But in my case, we were sending a burst of traffic while with a lower HTB bw limit, so we can consistently see backlog!=0 for many seconds. > > Even though your explanation above is inaccurate I think the spirit > of the patch looks reasonable. i.e keeping track of all additions to > the queue and removals from the queue in the backlog stats is useful. > However, you need to be extremely careful: This should only be done > at exactly the spot the packet is enqueued (and not by a parent's > enqueue asking for hierarchical enques). The reason why I care about backlog and qlen is I want to know the average length of each packet in backlog, to check if it is a GSO packet at least. > > I think some more work is needed Cong for this general patchset. > Sure, I could miss something somewhere, just point it out. :) Thanks. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html