Re: vmstat's entries type

M. Warner Losh wrote:
 > One approach that we could use for 64-bit counters would be to just
 > use 32-bits one, and poll them for overflow and bump an overflow
 > count.

That's an interesting idea.  That would eliminate the
efficieny concern.

 > This assumes that the 32-bit counters overflow much less often
 > than the polling interval,

Let me try some math ...  The context-switch counter on a
3+GHz machine here increases by 500 per second when idle,
and by 26,000 per second when under some serious load (so
it overflows after less than two days!).

Lets say we need to cope with 1,000,000 per second on
average, which would be really a lot.  In that case a
32bit counter would overflow after 1 hour 11 minutes,
which I think is plenty of time to poll for an overflow.
If we poll every five minutes, it would be sufficient
for an average of 14 million counter events per second.

 > and easily triples the amount of storage for each of them...

True, storage is tripled.  But how many counters are we
talking about here?  I guess rather a few, not thousands
of them, right?

 > It is ugly :-(

Yes, I certainly agree, it's ugly.  But having wrong output
from "vmstat -s" and other tools is ugly, too -- and it is
noticed by a lot more people.

Best regards
   Oliver

-- 
Oliver Fromme,  secnetix GmbH & Co. KG, Marktplatz 29, 85567 Grafing
Dienstleistungen mit Schwerpunkt FreeBSD: http://www.secnetix.de/bsd
Any opinions expressed in this message may be personal to the author
and may not necessarily reflect the opinions of secnetix in any way.

"Perl will consistently give you what you want,
unless what you want is consistency."
        -- Larry Wall