bus benchmark

[lennart@xxxxxxxxxxxxxx (Lennart Poettering)]

On Di, 27.03.18 22:24, zhiting zhu (zhitingz at cs.utexas.edu) wrote:

> Hi,
> 
> I have a question about the bus benchmark. On the v238, the benchmark only
> works with the input chart direct. If I put another input, the program gets
> an assertion failure. Also, what chart, bisec, legacy, direct means?
> There's no documentation come with the benchmark. I wonder what that output
> means? What metric does it try to show? Is this a latency or throughput
> benchmark?

That tool was written in order to determine when transport via memfd
would be faster than transport via AF_UNIX datagrams. i.e. the
datagram size at which point zero-copy transport (which implies
mmap()ing) becomes more efficient than copy-based transports.

If you so will it's a remnant of the kdbus experiments, and we should
probably remove it from the tree.

Lennart

-- 
Lennart Poettering, Red Hat