G'Day Rick,
On Thu, May 22, 2008 at 02:51:52PM -0500, rickey c weisner wrote:
> Brendan,
> I hope you get nicstat as well as your entire toolkit included in the
> OS distribution. Netstat alone does not cut it.
>
> As for NX it is my goal to have a lightweight tool that pretty much
> gathers most things related to the network that I might be interested
> in.
Ok; it might make sense to maintain that code as a download from your
blog (do you have a blog?). You can still help us plan what generic tool
could be included in Solaris for customers - whether it ends up based on
nicstat, nx.se, or written from scratch entirerly. In fact, I suspect
an argument could be made to have it as:
dladm iostat
A style that zpool uses. dladm has access to most details, and knows
which links are up and what their speeds are. Since it is already a
supported tool and is maintained, it may make the best place to build
from (something more than just dladm show-dev -s -i 1). We'd need to tell
the base of nicstat/nx.se users that their functionality can be found as
sub-options under "dladm iostat".
Anyhow, "dladm iostat" may not end up the best way to do it - it's another
option. As a stand-alone tool, nicstat is able to run on Linux, and if we
put the brains behind choosing the best metrics and documenting them, then
it could become an standard across different OSes.
> nx gives me tcp and udp as well as NIC related statistics.
> right now I get:
> 14:43:51 Iseg/s Oseg/s InKB/s OuKB/s Rst/s Atf/s Ret% Icn/s Ocn/s
> tcp 0.0 0.0 0.00 0.00 0.00 0.00 0.0 0.00 0.00
> IDGS/s ODGS/s IERR/s
> udp 0.0 0.0 0.00
> 14:43:51 Ipkt/s Opkt/s InKB/s OuKB/s IErr/s OErr/s Coll% NoCP/s Defr/s
> lo0 0.0 0.0 0.00 0.00 0.000 0.000 0.0 0.00 0.00
> e1000g0 0.0 0.0 0.00 0.00 0.000 0.000 0.0 0.00 0.00
>
> I plan to add listendrops in the next incantation.
>
> On my box
> nicstat gives me
> arwen:nicstat 1 1
> Time Int rKb/s wKb/s rPk/s wPk/s rAvs wAvs %Util Sat
> 14:45:24 e1000g0 0.52 0.17 0.65 0.51 819.31 331.07 0.01
> 0.00
> 14:45:24 mac 0.52 0.17 0.65 0.51 819.31 331.07 0.01 0.00
>
> all good stuff but I want the nocanputs in particular.
nocanputs may be a good metric for you, but it may not be good as a primary
metric for customers. The primary metric that nicstat shows is "Saturation"
which includes nocanputs. If a customer does see saturation, then a verbose
option could list extended details - all the metrics that can cause
saturation, including vendor and implementation specific ones such as
nocanput. If nocanputs becomes something else in the future (or is different
on a different OS), then we aren't breaking the primary default metrics.
nocanput is odd anyway, I'd be interested to hear what you do with it
and which interface types you've seen it on. :) I noticed a drop in visible
nocanputs between Solaris 9 and Solaris 10 (I couldn't prove why since I
didn't have source access at the time). Should I assume that my interfaces
are performing better, and that I no longer have a saturation issue? Or did
the implementation change, and it is no longer bumping the nocanput kstat,
but bumping something else instead?
I believe nocanput comes from the Solaris streams implementation, if
!canputnext() to a queue. Lookup canputnext in cscope/cvs.opensolaris.org -
some bump nocanput, many don't. In addition, many streams calls are skipped
these days in TCP/IP to improve performance. nocanput isn't value-less,
certainly if you see it then you know what is happening for that interface,
but as a default output it could be misleading. Your default output had
lo0 and e1000g0 - do either of them bump nocanput in Solaris 10?
cheers,
Brendan
--
Brendan
[CA, USA]
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
