Re: [Bug runtime/2514] New: Sequence had xxx drops.

[Martin Hunt <hunt at redhat dot com>]

On Wed, 2006-04-05 at 14:31 +0000, guanglei at cn dot ibm dot com wrote:
> One reason I guess is that in such a heavy loaded system, the probe handlers are
> triggered in a very high freq and generates too much data to be handled by
> relayfs. The speed of putting into relayfs transport layer is much faster than
> the speed that those data are retrieved away from relayfs from user space.
>
> So are there any ways of fixing this? It is a problem for the trace tool like
> LKET which will generate a very large trace data file.

The upper limit of both procfs and relayfs is the speed of the hard
disks, typically around 40MB/sec. If you are trying to log more data
than you can write to disk, you have a problem that cannot be fixed
without upgrading your hardware.

For bursts of data that temporarily exceed 40MB/sec, you can use "stap
-s" to allocate a larger buffer size.

It is possible to slow down the write process when the output buffer
starts getting full to avoid dropped packets. I have experimented with
this in procfs. However, that slows down the system and mainly just
reduces the need for large buffers. It cannot increase the sustained
logging rate which is limited by the drive speed.

Martin