Translating with DrWatson… this can take a few seconds the first time.
This is a genuine, complex translation. DrWatson protects commands, error codes, and log output while naturally translating the surrounding text. It’s translated once and saved.
InfoJones wrote:
>
> How am I certain that I don't have LRU contention going on?
>
> I understand that onstat -g spi can help - not sure how to read it
> though
> Will lru contention cause the bufwaits from an onstat -p to increase?
[output SNIPPED]
You can sometimes gleen LRU contention by looking at the number of
spins per loop in the -g spi output, yes. This value should be low
for a well tuned system. The best indicator of LRU and buffer
contention is what I have called the Bufwaits Ratio which is:
BR = bufwaits / (dskreads + bufwrits) * 100
Ideally this ratio should be around 7%, values over 10% are sure death
and a quiet system or one with no natural contention will show ratios
below 5%. In between 7 and 10% is a no-man's land where you will get
intermittent reports of slow performance for short periods.
To determine whether buffer contention or LRU contention is involved
try to calculate your buffer turnover rate:
Buffer Turnovers = (dskreads + bufwrits) / BUFFERS
BTR = (Buffer Turnover) / (time since last stats clearing) expressed
as turnovers per unit time.
If you are turning over your buffer pool more often than a few times
an hour you probably need more buffers. Add buffers and check again.
--
Art S. Kagel & Family
kagel@erols.com
Art S. Kagel & Family <kagel@erols.com> wrote:
: InfoJones wrote:
: [output SNIPPED]
: To determine whether buffer contention or LRU contention is involved
: try to calculate your buffer turnover rate:
: Buffer Turnovers = (dskreads + bufwrits) / BUFFERS
: BTR = (Buffer Turnover) / (time since last stats clearing) expressed
: as turnovers per unit time.
Should one multiply BUFFERS by page size in the Buffer Turnovers calculation
or go with just the value of BUFFERS from the config file?
: If you are turning over your buffer pool more often than a few times
: an hour you probably need more buffers. Add buffers and check again.
: --
: Art S. Kagel & Family
: kagel@erols.com
--
bothwellno@spam.duke.edu
We use strictly necessary cookies to make this site work. With your
consent we’d also use optional cookies for analytics and marketing. You can accept all,
reject all, or choose. Read our Cookie Policy.