On 07/11/2013 02:36 AM, Erwan Velu wrote:
On 10/07/2013 18:01, Mark Nelson wrote:
Hello again!
Part 2 is now out! We've got a whole slew of results for 4K FIO tests
on RBD:
http://ceph.com/performance-2/ceph-cuttlefish-vs-bobtail-part-2-4k-rbd-performance/
Hey mark,
I'm really fond of this kind of plotting performance results. You may
saw my work on this topic that is about to be upstreamed in fio :
http://www.spinics.net/lists/fio/msg02140.html (one tool to automatize
the fio jobs, one for graphing). You may find some interest in.
My only comment about thoses graphs will be maintaining comparable
ranges across same benchmark types.
For example, on this picture,
http://ceph.com/wp-content/uploads/2014/07/cuttlefish-rbd_btrfs-write-0004K.png,
we do have 6 runs with the same IO profile but each result is plotted in
its own context.
Meaning the Zaxis range and its associated colorbar (CB) is changing all
the time. That makes very difficult to compare runs as the color doesn't
have the same meaning everywhere.
I'm usually enforcing both zrange & cbrange (looks like you use gnuplot)
to be constant over graphs.
Syntax is : (to be adjusted in values)
set cbrange [0:120000]
set zrange [0:120000]
Pros:
- the same color have the same raw value everywhere
- a quick look at a graph series gives immediate an understanding of the
overall performances (which one is good, which one is bad)
- speed up graph reading
Cons:
- it could visually flatten some low values
- needs to estimate the range before plotting
My 2 cents,
Erwan
Hi Erwin,
Yes I agree 100%! I wanted to do that, but ran out of time. I think
(as you stated) the only way to pull it off is to globally figure out
the maximum z-axis range for the group, then manually pass it in to
gnuplot for the graph generation. Unfortunately when I was creating the
plot file I wasn't planning at the time to glue the graphs together with
montage, so I didn't really think to design it to work that way. I'm
discovering that I probably need to rely less on doing things internally
in the plot file and more on wrapping the plot generation in something
else like python.
Your graphs look very similar to the ones I made! You'd almost think we
did it together. :) I see you are also running into the same
overlapping labels issue that I've hit:
http://www.flickr.com/photos/ennael/9101313574/in/set-72157634249027122
Anyway, glad someone else is as crazy as I am for wanting to graph so
much data. ;)
Mark
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
