Probably needs some tidying up before it's mergeable, but I like it.
Is useful as is, so I thought other people might want to look at it and
give me feedback. Based very loosely on a earlier patch by Andrew Theurer
that gave nr_running per cpu, but I wanted to see load averages. This is
extremely useful for looking at scheduler rebalancing problems.
I think we should be basing the scheduler rebalance calculations off load
averages, rather than a couple of sample points - I'll create a patch to do
this sometime soon.
The normal load average calculates nr_running() + nr_uninterruptible()
which can be confusing. real_load_average just uses nr_running.
Output here is from a NUMA 16-way machine running make -j256 bzImage on the
kernel.
larry:~# cat /proc/real_loadavg
Domain load1 load2 load3 nr_run/nr_thrd
SYSTEM 80.67 25.75 9.02 253/901
0 5.09 1.61 0.56 18/901
1 5.17 1.66 0.58 17/901
2 5.28 1.70 0.59 14/901
3 4.80 1.54 0.54 17/901
4 5.39 1.72 0.60 16/901
5 5.17 1.66 0.58 15/901
6 4.72 1.50 0.53 15/901
7 4.62 1.46 0.51 16/901
8 5.27 1.69 0.59 18/901
9 5.06 1.61 0.56 15/901
10 5.15 1.63 0.57 16/901
11 5.06 1.59 0.55 15/901
12 5.45 1.70 0.59 17/901
13 4.77 1.56 0.54 15/901
14 4.89 1.55 0.54 15/901
15 4.74 1.50 0.52 14/901
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Feb 15 2003 - 22:00:57 EST