Re: HyperThreading

[Saman Barghi <saman.b@xxxxxxxxx>]

It would be great if you had the SMP results as well. Anyway, I think you would love this:
 http://www.cs.uwaterloo.ca/~sbarghi/bench/transactions1.jpg
(keep in mind that this is a quad core xenon with 2 GB of RAM, and FreeBSD is in SMP mode)

As I said before; I kept the environment the the same as what has been mentioned here: http://people.freebsd.org/~kris/scaling/dfly.html . I even used the same version of MySQL (5.0) although MySQL  5.5 shows better scaling behaviour.(http://mikaelronstrom.blogspot.com/2010/04/mysql-554-m3-scales-to-32-cores.html). IMHO if "The multiprocessor work that has been ongoing in DragonFly is really starting to bear fruit", the fruits are indeed juicy and sweet.  It would be lovely to see the results on a machine with more cores though.

Btw, is there any changelog that shows what has been changed about tokens !?


Cheers,
Saman

On Fri, Feb 18, 2011 at 11:00 AM, Venkatesh Srinivas <me@endeavour.zapto.org> wrote:

Sysbench OLTP is a very solid test; I'd love to see any results you get from it, either with Postgres or MySQL as a database server.

http://m-net.arbornet.org/~sv5679/sysbench_nmalloc_df.gif are results from the spring of last year, very early in the 2.7 branch, for Sysbench/OLTP with MySQL against Dfly. Despite the title of the graph, this was on an 8-core Xeon (Nehalem-based) system; also not shown on the graph are the UP kernel results -- the UP kernel with the old libc malloc had a more or less flat line at 650 transactions/sec (which was distressingly always faster than the SMP kernel!) and the UP kernel with the new libc malloc had a flat line at around 800 transactions/sec. With Google's tcmalloc, we had much better results, but I don't have those numbers handy.

We've really broken up the MP lock since then; the LWKT scheduler has been changed a lot (round-robin -> fair share); tokens have been reworked twice; and even the 4BSD scheduler has had work. I'd love to see if we're merely rearranging deck chairs or if we've actually improved matters...

-- vs