[IQUG] [iqug] LMA usage used vs allocated

John Anthony chirayithaj at gmail.com
Thu Mar 29 13:20:13 MST 2018


Here is how i understand it from a year ago but there has been substantial
work in this area , so others with their ears in right places should be
able to clarify!

Like all IQ memory, not all the allocation is made at startup, but once
allocated it will likely not be released, even if the current usage does
not require the memory. Given that, "IQ Large memory inflexible" is
typically much below -iqlm configuration, which also indicates waste of
memory on very large deployments, something we have discussed here in the
recent past.

However, "IQ Large memory inflexible" approaches -iqlm then there is
potential churn (swapping) within the large memory space. But it is also
possible that this was a one off spike for LMA and this happen very often
given how LMA is configured. What is required is some way to determine if
there real churn in LMA memory.

N-Bit processing takes a signficant hit if pages are not in memory, on the
one hand, but on the other, on very large systems the best practice
recommendations are wasting memory.

Also, with very large systems, the default FP_NBIT_AUTOSIZE_LIMIT value is
impractically small, and there needs to be experimentation with values in
the 1 billion + range. None of this can happen without precise metrics on
LMA usage profile.

Bottom line - we need a usable method to collect the metrics to answer this
question.

-JA

On Thu, Mar 29, 2018 at 2:34 PM Louie, David <David.Louie at blackrock.com>
wrote:

>
> I'm not finding any good doc on how LMA works.
>
> On some systems I see LMA allocated vs used with large differences
>
> [cid:image001.png at 01D3C773.51E5AF90]
>
> In this case there is no demand for LMA based on the load and query
> activity in this database.
>
> However in some servers the allocated vs used is very close but never
> encounters any our LMA issues:
>
> [cid:image002.png at 01D3C773.51E5AF90]
>
>
> So my question is "IQ Large memory inflexible" used can grow to close to
> the "IQ Large memory space"  (I've seen the inflex used reach up to within
> 1 mb of the allocated)
> is IQ swapping out memory from the used (using LRU MRU logic) to
> accommodate other processes which need the memory behind the scenes but we
> are not just seeing it?
> (I killed all processes on a server where inflexible used and allocated
> were close in values and didn't see usage go down)
>
> David
>
>
>
> This message may contain information that is confidential or privileged.
> If you are not the intended recipient, please advise the sender immediately
> and delete this message. See
> http://www.blackrock.com/corporate/en-us/compliance/email-disclaimers for
> further information.  Please refer to
> http://www.blackrock.com/corporate/en-us/compliance/privacy-policy for
> more information about BlackRock's Privacy Policy.
> For a list of BlackRock's office addresses worldwide, see
> http://www.blackrock.com/corporate/en-us/about-us/contacts-locations.
>
> (c) 2018 BlackRock, Inc. All rights reserved.
> _______________________________________________
> IQUG mailing list
> IQUG at iqug.org
> http://iqug.org/mailman/listinfo/iqug
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://iqug.org/pipermail/iqug/attachments/20180329/919858d5/attachment.html>


More information about the IQUG mailing list