This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: Where stap spends its time producing instrumentation for testsuite/systemtap.syscall/sys.stp
- From: fche at redhat dot com (Frank Ch. Eigler)
- To: <systemtap at sources dot redhat dot com>
- Date: Wed, 03 Oct 2007 23:31:10 -0400
- Subject: Re: Where stap spends its time producing instrumentation for testsuite/systemtap.syscall/sys.stp
- References: <4703E728.9000408@redhat.com>
wcohen wrote:
> [...] Below is the the experiment to see how long it takes to build the
> [systemtap.syscall/sys.stp] instrumentation with oprofile collecting
> data.
Thanks.
> The analysis pass (pass 2) and compilation pass take a over 30
> seconds each.
Yeah, that's far too long.
> Pass 1: parsed user script and 37 library script(s) in 530usr/20sys/626real ms.
> Pass 2: analyzed script: 768 probe(s), 108 function(s), 14 embed(s), 4
> global(s) in 39690usr/360sys/40385real ms.
This part is likely to improve with the better elfutils-dwfl caching
widget I'm working on for the user-space probing support.
> Pass 3: translated to C into "/tmp/stapnoy9IG/xyz.c" in 300usr/110sys/462real ms.
> Pass 4: compiled C into "xyz.ko" in 34080usr/1500sys/36972real ms. [...]
> $ wc /tmp/stapnoy9IG/xyz.c
> 120522 352489 4295506 /tmp/stapnoy9IG/xyz.c
> $ size /tmp/stapnoy9IG/xyz.ko
> text data bss dec hex filename
> 1134094 169408 4816 1308318 13f69e /tmp/stapnoy9IG/xyz.ko
Both compilation time & result size numbers will have improved greatly
since the time you took these measurements due to some code generation
improvements (see bug #5096 and #5102).
There is a chance that future implementation of bug #5075 could create
many more code elision opportunities as $argstr computation is moved
out of explicit script code.
- FChE