Re: "Strange" values in Monitor in a GS80 Galaxy configuration
From: Carl Perkins (carl_at_gerg.tamu.edu)
Date: 06/03/03
- Next message: Bob Koehler: "Re: Intel 64 bit Pentium seems to be on its way"
- Previous message: BoylesA: "Re: ISACFG on AlphaStation 500"
- In reply to: Roy Omond: ""Strange" values in Monitor in a GS80 Galaxy configuration"
- Next in thread: John Brandon: "Re: "Strange" values in Monitor in a GS80 Galaxy configuration"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Date: 3 Jun 2003 07:42 CDT
Roy Omond <Roy@Omond.net> writes...
}Gentle colleagues, one of my clients has posed the following
}Monitor query to me. This is on VMS 7.3-1, well up-to-date with
}patches, on a GS80 in a Galaxy configuration. Node XE has one
}CPU migrated to it (I'm pretty sure it starts off as one CPU,
}i.e. it gets a second CPU, though I can't be certain exactly when
}the CPU was migrated, but I'm assuming it was during the Monitored
}period), and this seems to, ahem, "confuse" Monitor as seen below.
}
}Is this expected behaviour (I understand it might be rather difficult
}to get this right in all circumstances) ? Any comments ?
}
}Thanks in advance,
}
}Roy Omond
}Blue Bubble Ltd.
}
}--- cut here ---
}
}We run a daily monitor job between 14:00 and 15:00 using this command:
}
}$ MONITOR PROCESS/TOPCPU,SYSTEM/ALL/NODISPLAY-
}/INPUT=filename/SUMMARY=filename/BEGINNING=14:00/ENDING=15:00/NODE=noden
}ame
}
}We are running a GS80 containing two nodes - XE and XN. If we move
}one CPU to XE the above monitor job running on XE gives the
}following results:
}
} OpenVMS Monitor Utility
} SYSTEM STATISTICS
} on node XE From: 28-MAY-2003 14:00:09
} SUMMARY To: 28-MAY-2003 15:00:09
}
} CUR AVE MIN MAX
}Interrupt State 8.90 108.68 3.69 1210.23
}MP Synchronization 1.12 102.52 0.00 1220.00
}Kernel Mode 11.76 1041.19 5.23 12368.69
}Executive Mode 2.15 189.44 0.73 2251.14
}Supervisor Mode 0.41 9.30 0.19 108.20
}User Mode 16.96 1579.34 8.47 18745.42
}Compatibility Mode 0.00 0.00 0.00 0.00
}Idle Time 158.66 12923.51 77.38 153655.51
}
}We are especially interested in the Idle Time figures but the ones I've
}highlighted above are a little strange to say the least.
}
}Is this a bug? If so, is there a fix? If not, can you explain these
}figures to me?
I would say that when you add a CPU *while* monitor is running that it
may not have some arrays or such set up properly so you pick up junk.
It is also possible (probably fairly likely even) that it was a single
bad sample that happened when the processor was added. If so, you
should see that single bad sample in the CUR column at some point -
I would guess that the bad sample will pretty much match the values
in the MAX column as they are all over 100, which shouldn't happen
when the instance is single processor.
My guess of what happens:
MONITOR is cruising along checking the time in each processor mode
for each CPU and adding them up. This is easy when there is only 1 CPU.
Then a second CPU magically appears. The CPU tick counts stored in the
kernel are absolute counts - if you add them all up and you get the time
since boot (in "ticks" - with an Alpha each tick in the count appears to
represent 1/1024 of a second, on a VAX they used to be .01 seconds).
So suddently there is this second CPU with a count. To get the time in
each mode in the interval, you take the current value (a 64 bit integer)
and subtract off the value from the previous sample and then multiply it
by the "tick" to "seconds" conversion factor. Except there was no "previous
sample" - it is 0 for each if the memory location it checks for this exists
and was initialized to 0, otherwise it is whatever random junk was in the
location it checks. So you end up with either the total number of CPU ticks
the processor has had in each mode, or that number minus a randomish number,
instead of the number of ticks since the previous check.
The system wouldn't happen to have been up for about 2.19 days when the
second CPU was added, would it? If so, then the new processor probably
had it's initial CPU tick counters set to match the first processor's
(or MONITOR sees it as such) when it was added, or something along those
lines. (If you add up the MAX column's times they add up to a little
over 2 days, 4 hours, 39 minutes). I ask because the numbers may not
just be random junk, they might be junk with some (undesired) meaning.
--- Carl
- Next message: Bob Koehler: "Re: Intel 64 bit Pentium seems to be on its way"
- Previous message: BoylesA: "Re: ISACFG on AlphaStation 500"
- In reply to: Roy Omond: ""Strange" values in Monitor in a GS80 Galaxy configuration"
- Next in thread: John Brandon: "Re: "Strange" values in Monitor in a GS80 Galaxy configuration"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]
Relevant Pages
|
