GNOME Bugzilla – Bug 518355
misleadingly bursty network transfer rate
Last modified: 2011-11-11 10:03:55 UTC
System Monitor's Network History graph makes my transfer look misleadingly bursty. I have a high transfer rate on one update, then no data at all on the next, then back to high again, and so on. Every other update shows zero bytes transferred. This does *not* agree with that actual arrival of data, which is much more smooth. Unfortunately it makes it very hard to see what my real transfer rate is: the reported "Received:" rate also bounces back and forth between 0 and some much larger value. I am currently using the default update interval: 1 second. If I increase that to 1.5 seconds, then every fourth update is zero instead of every other update. If I increase the update interval to 1.75 seconds, then the zeros go away entirely. But simply slowing things down is not entirely satisfying. I'd rather see reasonable updates at the reasonable default 1-second update interval. Note that this problem affects both the System Monitor application and the panel applet.
Could you run something like: for i in $(seq 1 10); do cat /sys/class/net/eth0/statistics/rx_bytes; sleep 1; done while downloading ?
Output from "repeat 10; do cat /sys/class/net/eth0/statistics/rx_bytes; sleep 1; done" is as follows: 1780775657 1780775657 1781426867 1781426867 1782079547 1782079547 1782730757 1782730757 1783381967 1783381967 So the bursty updates are coming from the kernel itself. :-( Note that data arrives in the downloading application at a fairly smooth rate. I was seeing about 300K/s reported by my downloading application (wget) while the above numbers were being collected. So this is clearly not system-monitor's fault. But does anyone have a clue why rx_bytes would update so infrequently?
what about a tcpdump/tshark capture ?
tshark capture shows data arriving continuously and steadily. There are no 1-second gaps between packets. This is consistent with observed arrival of data by the downloading application. I reiterate my initial statement that data reception is not *actually* bursty.
Then you have to open a bug on http://bugzilla.kernel.org/
Opened as <http://bugzilla.kernel.org/show_bug.cgi?id=10600>.
*** Bug 532995 has been marked as a duplicate of this bug. ***