On Mon, Nov 26, 2018 at 10:13 PM Nathan Stratton Treadway <[email protected]> wrote: > > On Mon, Nov 26, 2018 at 17:07:54 -0500, Chris Hoogendyk wrote: > > My understanding was that Amanda cannot throttle the throughput, but > > it can see what it is. If it is at or over the limit, it won't start > > another dump. So, you can end up over the limit that you have > > specified for periods of time, but it won't just continue increasing > > without respect for the limit. > > To expand on this slightly (and tie it back to the begining of this > thread): what Amanda actually looks at is the dump-size/time figure from > the "info" database it keeps on each DLE. So, it's not looking at > dynamic real-time network bandwidth usage, but rather calculating what > the average throughput turned out to be the last time this DLE was > dumped. > > As you say, the important thing to note is that Amanda uses this info to > do throttling at the dump/dumper level: when there is a free dumper, it > will send a request for a new DLE to that dumper only if the bandwidth > usage calculated for currently-in-progress dumps doesn't exceed the > configured limit. > > If the configured limit is too high, the possiblity is that too many > simultaneous dumps, especially of very fast client systems, will > saturate the network interface (or some part of the network > infrastructure) -- while if it's too low, the symptom will be that some > of the configured "inparallel" dumpers will be left idle. (But the nice > thing about having Amanda do this calculation is that it can go ahead > and kick off more dumpers when working on slow client systems and fewer > when processing the fast ones.) > > Anyway, in the case of the original email in this thread, the problem > seems to be that the calculated bandwidth for the single DLE in processs > of being dumped already exceeds the configured bandwidth limit (as > represented by the amstatus report line "network free kps: 0") -- and > thus the other 9 dumpers are all left idle even though there are many > DLEs out there waiting to be dumped.
Nathan: Thank-you for the very clear explanation of how amanda handles this. That's good wiki material. So last night's run borked due to my having two netusage statements in the config. For some reason I missed that yesterday. Maybe its age-related, I don't know... ;-) That's fixed and we'll see what sort of throughput the network can handle tonight.
