Re: [External] Re: accumulo 1.10 tuning

["Ligade, Shailesh [USA]" <Li...@bah.com>]

Thanks!

-S
________________________________
From: Christopher <ct...@apache.org>
Sent: Wednesday, October 13, 2021 1:46 PM
To: accumulo-user <us...@accumulo.apache.org>
Subject: [External] Re: accumulo 1.10 tuning

I'm not aware of any existing prescriptive recommendations of the type
you are asking for. I think most recommendations tend to be reactive
and specific, rather than prescriptive and general, because so much
depends on the particulars of a user's situation. There are too many
variables, and each user's situation is different. What works for one
person's data and environment may not work for you in your
environment.

Tservers generally don't require a lot of heap for ingest. Make sure
you reserve enough room for the OS, and other processes on the
machine. And, don't forget to account for the native memory taken by
native compression libraries, like GZip. Monitor your tservers to see
how much heap you're using in your workload, and make sure you adjust
to optimize Java GC runs. Take into account your iterators and what
they are doing, as your iterators may require more memory. To optimize
your workloads, you may wish to experiment with running multiple
tservers with smaller memory footprints on the same server, rather
than a single tserver with a larger memory footprint. These are just a
few things to consider. Everybody's use case is unique.

On Wed, Oct 13, 2021 at 8:48 AM Ligade, Shailesh [USA]
<Li...@bah.com> wrote:
>
> Hello,
>
>
>
> I saw various guidelines on how to set memory heap sizes etc. IUs there a pluggable spreadheet like if the server has x memory, tserver heap should be x/3 and datanode should dhave x/4 etc?
>
> Also, is there any recommendation on number of tables hosted per tserver and tserver memory requirement? May be that will also provide  when to add a new tserver (when # of tablets per tserver goes above some threshold)
>
>
> -S
>
>