This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: Interoperability of LTTng and LTTV with SystemTAP (heavy usage)
- From: Martin Hunt <hunt at redhat dot com>
- To: Mathieu Desnoyers <compudj at krystal dot dyndns dot org>
- Cc: ltt-dev at shafik dot org, "systemtap at sources dot redhat dot com" <systemtap at sources dot redhat dot com>
- Date: Mon, 19 Dec 2005 14:06:22 -0800
- Subject: Re: Interoperability of LTTng and LTTV with SystemTAP (heavy usage)
- Organization: Red Hat Inc
- References: <20051214022515.GA1488@Krystal> <20051216010459.GA11668@Krystal>
On Thu, 2005-12-15 at 20:04 -0500, Mathieu Desnoyers wrote:
> Hi,
>
> Here is, as promised, the rest of the heavy usage scenario requested. It has
> been kindly generated by Autodesk partners.
> I would be interested to know if SystemTAP can handle an average load of 58K
> events per second.
A simple probe that logs a small amount of data typically takes <
5usecs, which is 200K events per second. I've seen 780K events on my
dual-processor Xeon. However, that uses almost all the cpu time.
> Heavy usage scenario : Trace from Autodesk, 64 bits x86_64, 4 CPUs
>
> 621 MB trace. Duration : 9m10s.
That's only 1.1MB/sec. With relayfs or procfs, I get 40MB/sec until the
buffers are filled, with sustained rates between 10-20 MB/sec depending
on disk type.
> Time used by lttd :
>
> Percentage of wall time :
> 3.143238573/2185.918765540 * 100% = 0.144%
> Percentage of cpu time :
> 3.143238573/1050.873236386 *100% = 0.299%
>
>
> overall event count (for the whole 9m10s on the 4 CPUs)
>
> events count : 31946639
For systemtap: Worst case for a small probe, 5usec/event, so that's
160 sec cpu time.
Percent of wall time: 7.3
Percent of cpu time: 15.2
That's worst case for a simple trace. I actually benchmark kprobes at
1.4usec and a systemtap script that does a printf of a small string at
2.5usecs. So a best case would be half the above numbers.
Martin