A damp discussion of network queuing
A damp discussion of network queuing
Posted Oct 17, 2014 17:17 UTC (Fri) by nix (subscriber, #2304)In reply to: A damp discussion of network queuing by mtaht
Parent article: A damp discussion of network queuing
There are some exceptions, notably a good hi precision clock source is needed.IIRC, fq_codel also needs BQL support in the NIC driver. Some embedded firewall boxes (in my case, the Soekris net5501) have NICs such as the VIA Rhine for which BQL is not implemented yet. (There are old patches for the Rhine, but nothing for recent kernels that I know of.)
Hm actually I think you mentioned this a few months ago on the cerowrt list. Great minds think alike etc etc.
Posted Oct 17, 2014 19:09 UTC (Fri)
by mtaht (subscriber, #11087)
[Link] (1 responses)
http://snapon.lab.bufferbloat.net/~d/beagle_bql/bql_makes...
I do hope that the more companies realize that BQL support is essential to high performance (I'm looking at *you*, Arm, Cisco, AMD, and Xilinx and a dozen others), the more BQL drivers (with xmit_more support) will land on everything. Certainly nearly all the 10GigE makers "get it", but that knowledge has not fully propagated down into the older and slower devices...
http://www.bufferbloat.net/projects/bloat/wiki/BQL_enable...
There is a paper in progress on how much BQL helps - answer, quite a lot - while we (in the bufferbloat world) know this, that sort of stuff needs to land on CTO and academic and driver developer desks.
I wrote up some issues are in adding BQL to a device driver here, I had planned to write a tutorial but haven't got around to it.
https://lists.bufferbloat.net/pipermail/bloat/2014-June/0...
So far as I recall the via rhine was updated to BQL recently, but will check.
Posted Oct 21, 2014 16:25 UTC (Tue)
by nix (subscriber, #2304)
[Link]
A damp discussion of network queuing
A damp discussion of network queuing