bufpages kernel parameter on 11.23 11iv2

[support_5]

Hi,

I notice that in kcweb, there is no longer an option to set the size of the buffer cache on 11.23. How can we statically set the size of the buffer cache and turn off dynamic buffer cache? I ask because dynamic buffer cache is crap.

ie if you leave it at the default of 10% (max_dbc_pct) then on a 20GB Ram system, you have 2GB of buffer cache which is ridiculously high (you rarely need any more that 400MB).

We used to set buffer cache to 600MB statically, but now the only way we can do that is by making dbc_max_pct and dbc_min_pct both set to "3" or "4", but it's not a very granular level of control. Why has bufpages been taken out of the kernel parameters?

Anyone know?

Thanks heaps.

- Andy Gray

[James R. Ferguson · Accepted solution]

Hi Andy:

Have a look at Technical Knowledge Base document #UKRNBKBRC00016993. Both 'nbuf' and 'bufpages' _are_ available in 11.23. Use 'kctune' to tune them.

Regards!

...JRF...

[Kent Ostby]

Andy -- of course you can go the dbc_max_pct route as you mention.

600 mb is good max number although going a bit higher (say 800 mb) isn't going to hurt you either.

[support_5]

I wonder why it was hidden? Why would they have hidden those parameters? Are they trying to get everyone to use the dynamic buffer cache? If so, why not improve it so that it works better? Dynamic buffer cache, as I have mentioned is really hopeless! It always hovers around the maximum mark even when there is significant memory pressure. So yeah, useless.

- Andy

[Ted Buis]

Dynamic buffer cache can be useful for some environments (often technical computing ones), but I would agree that the general rule is that it doesn't do much for most applications with truly random I/O to data that is much much larger than the cache. One might also question why it is the default or why there isn't a better algorithm for picking more intelligent values for the buffer cache size.

[support_5]

Yes, I agree Ted. Oh well. Thanks for the help fellas.

- Andy