| Newsgroups |
gmane.comp.sysutils.pcp |
| Message-ID |
<[email protected]> |
On Mon, Dec 12, 2016 at 5:31 PM Marko Myllynen <[email protected]> wrote:
> Perhaps the other Marko who raised the issue could clarify a bit his
> main concerns with visualization but from what I've seen, visualization
> on non-Linux platforms among larger teams is the number one issue
> preventing PCP's further adoption. If there are no shortage of developer
> cycles on this front then it's of course perfectly fine to develop both
> but as a project I think PCP should today emphasize web based
> visualization tools.
>
Let's specify what I am talking about:
1. Topic starter talked about 3 thousand servers. We are collecting data
from ~300 servers.
2. We would like to collect metrics once per second. And collecting per
proc metrics is also something that could be very useful for us.
3. We think that collecting this data for 2 or 3 day would be enough. If
you need data for longer, just copy specific archive for later use.
Currently available visualization tools have problems dealing with such
amount of data. Especially if you collect per proc metrics.
I haven't touched pcp in a while because of a lack of time, unfortunately,
but as I recall:
- pmwebd barely works on such amount of data, some operations can take
minutes with 100% usage
- pmchart is faster, but it's not very productive tool, unfortunately
I've already raised few such issues here and in bug tracker.
Right now pcp daemon works on our servers and you can use it to see live
data or collect something manually.
But collection and visualization from all servers to dedicated machine is
in a broken state because of these problems.
I understand that these things that I've said do not help. You, developers,
need specifics to fix all the problems that we have. And I plan to give
them to you as soon as I manage to find some time to return to this project.
But big picture is that pcp, as good as it is, is not very usable for us
right now.