3
* Based on ftrace. Not always checking existance of all trace points. Therefore, may
4
show system idle because of missing tracepoint in the kernel:
7
perf-stat-hist net:net_dev_xmit len 10 - Histogram of used packet sizes
8
perf-stat-hist syscalls:sys_enter_read count 10 - Histogram of issued read request sizes
9
iolatency - Histogram of io latencies
11
iosnoop -ts [-Q] - Latencies of disk accesses with issuing apps (from time of issue and from time of queue)
12
opensnoop - Trace file opens (by app)
14
tpoint -H block:block_rq_insert - Just trace a specific tracepoint
15
tpoint -s block:block_rq_insert 'rwbs ~ "*R*"' - Only reads and show stack trace
16
kprobe -H 'p:do __do_page_cache_readahead nr_to_read=%cx' - kprobe
18
Kernel function analysis:
19
funccount -i 1 submit_bio - Counts how often specific kernel symbol was executed (per 1 second)
20
funcslower -P filemap_fault 1000 - Reports when the specified kernel function takes more than 1000us to execute
21
funcgraph -P filemap_fault - Show function execution tree
24
... Other useful tools for cache, io, etc. ...