Hi, 2015-03-16 19:23 GMT+01:00 David Lang <[email protected]>: > On Mon, 16 Mar 2015, Mickael Chazaux wrote: > >> If no HR timer is present, then the only time source is the Linux OS >> tick at (typically) 100Hz. > > > is OpenWRT compiling with 100Hz Jiffies?, 1000Hz has been available for a > long time.
I had this problem with a custom-configured upstream 2.6.32 kernel on an Atmel ARM SAM9G20 CPU. Putting the tick at 1kHz helped, but the TBF still was not precise. The fix was to enable HR timers. It was quite a few years ago, so I can't recall all the details, I believe I investigated the issue by looking at the TBF code, then looked at the delay/sleep/timer functions, then fiddled with the HZ value before I found out about the HR timers, then enabled them. > >> If a shaper decides to put a 10 ms delay >> between each of your typical 1500byte ethernet frame, it means max >> 1500*8*100 = 1.2Mbit/s egress. > > > the shaper should not be adding a constant delay between packets, so I don't > think it's quite as bad as you are thinking. > It was a few years ago, I don't remember everything. The symptoms described do ring a bell though. Mickael _______________________________________________ openwrt-users mailing list [email protected] https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-users
