Re: [pcp] Floating point problem

[Martin Spier <mspier@xxxxxxxxxxx>]

Here it is:

kernel.pct.cpu.user = 100 * kernel.all.cpu.user / hinv.ncpu

kernel.pct.cpu.sys Â= 100 * kernel.all.cpu.sys / hinv.ncpu

Same definition Amer posted before. Think it came from:

http://www.performancecopilot.org/pcp.git/man/html/howto.cpuperf.html

Here are two sample responses I just took, 1 second apart:

{

 "timestamp": {

  "s": 1406575780,

  "us": 397728

 },

 "values": [

  {

   "pmid": 2143289345,

   "name": "kernel.pct.cpu.user",

   "instances": [

    {

     "value": 5.35574E+8

    }

   ]

  }

 ]

}

{

 "timestamp": {

  "s": 1406575781,

  "us": 393553

 },

 "values": [

  {

   "pmid": 2143289345,

   "name": "kernel.pct.cpu.user",

   "instances": [

    {

     "value": 5.35574E+8

    }

   ]

  }

 ]

}

Since I'm calculating deltas, the precision only gives me full percentage points at this stage. In this case, since the system was idle, it was either 0 (like the response samples) or 1.

â

Let me know if there's anything wrong with the definitions.

Thanks,
Martin

On Sat, Jul 26, 2014 at 12:40 AM, Ken McDonell <kenj@xxxxxxxxxxxxxxxx> wrote:

I think there is a problem with the derived metric definition â looks like the number is a counter, which probably is not correct.

Â

Can you send me the derived metric definition please?

Â

From: pcp-bounces@xxxxxxxxxxx [mailto:pcp-bounces@xxxxxxxxxxx] On Behalf Of Martin Spier
Sent: Saturday, 26 July 2014 10:59 AM
To: pcp@xxxxxxxxxxx
Cc: Amer Ather; Coburn Watson; Brendan Gregg
Subject: [pcp] Floating point problem

Â

Ran into a small problem with "value" floating point precision on pmwebd json responses.

Â

Cumulative values get too big and I start having problems like this:

Â

{

 "timestamp": {

  "s": 1406323733,

  "us": 288692

 },

 "values": [

  {

   "pmid": 2143289348,

   "name": "kernel.pct.cpu.user",

   "instances": [

    {

     "value": 7.94874E+10

    }

   ]

  }

 ]

}

Â

Since I'm trying to calculate % usage in this case, subtracting previous value from current and normalizing by delta in time, at this point I can't get any useful information to plot due to the low precision.

Â

Any suggestions?

Â

-Martin