pcp
[Top] [All Lists]

Re: [pcp] osvis/clustervis network percentages

To: tore.larsen@xxxxxxxxxxxxxx
Subject: Re: [pcp] osvis/clustervis network percentages
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue, 31 Jul 2012 10:35:00 +1000
Cc: pcp@xxxxxxxxxxx
In-reply-to: <OF5292E525.8FB8DBEF-ONC1257A4B.004F7ADD-C1257A4B.00510785@xxxxxxxxxxxxxx>
References: <OF5292E525.8FB8DBEF-ONC1257A4B.004F7ADD-C1257A4B.00510785@xxxxxxxxxxxxxx>
Sender: ndsco1@xxxxxxxxxxxxxxxxxx
Hi Tore,

On 31 July 2012 00:45,  <tore.larsen@xxxxxxxxxxxxxx> wrote:
> Hi,
>
>     pcp 3.6.3 pmcd, it seems everything is scaled to 100Mbps, rather than
> 1Gbps.  Where do I change pmcd to scale correctly on pmchart/osvis/pmview
> etc ?  Multiply ethX with 10, bond0 with 20 and ib with 400 (QDR) ?

Could this be a bits-vs-bytes confusion?  The pcp client tools will display
bytes, always, and never bits like network-centric tools tend to do.  This
makes sense from a system-level analysis point-of-view, where you might
be comparing bytes coming in to bytes being written to disk, or in memory,
etc ... but can confuse people who are used to focussing on network tools.
[ yes, Rodney, I'm looking at you ;) ]

If this doesn't explain the values you're observing, the place to look on the
server side is src/pmdas/linux/proc_net_dev.c and pmda.c - in particular,
the latter contains the metric definitions including the units.  Those are the
"master" definitions for individual metrics - the client tools only make use
of metadata from the server.

cheers.

--
Nathan

<Prev in Thread] Current Thread [Next in Thread>