Validate +stbt nnts as a potential default setting.

[evanmcc]

When running on bare-metal dual socket/cpu systems, +sbt (and stbt) nnts can improve performance. +stbt is preferred because it is ignored on platforms that do not support it.

In order to make it a default setting for vm.args, though, we need to verify a few things:

• performance isn't negatively impacted on other platforms (single cpu, platforms that don't support sbt, virtualized hardware)
• performance isn't negatively impacted on other backends (leveldb, memory, most of the testing has been on bitcask and memory so far, but signs for leveldb are good).

[jaredmorrow]

Marking this as 2.1, unless this is already fixed @evanmcc

[evanmcc]

IMO, we should just do it , or do +sbt db for 2.0-RC, but I have never had
the time to verify my intuition that it doesn't hurt machines with a single
numa domain.

[jaredmorrow]

Changing milestone to 2.0-RC, so we can come back and track this.

[jonmeredith]

@evanmcc @jaredmorrow I think the time window has closed on this, unless we're planning to rerun backend/cluster testing with this enabled.

[evanmcc]

I regularly recommend this setting to users with dual-core machines, and
thus run most of my scalability tests with this setting enabled. That
said, I don't typically run these tests on single core machines. Even
Kostis recommends it! But I am OK with being super-conservative for one
more release, I guess.

On Mon, Apr 21, 2014 at 8:36 AM, Jon Meredith notifications@github.comwrote:

@evanmcc https://github.com/evanmcc @jaredmorrowhttps://github.com/jaredmorrowI think the time window has closed on this, unless we're planning to rerun
backend/cluster testing with this enabled.

Reply to this email directly or view it on GitHubhttps://github.com//issues/391#issuecomment-40944687
.

[jaredmorrow]

I don't have a strong opinion on this one, though I wonder if the things mentioned in your first comment have been addressed?

In order to make it a default setting for vm.args, though, we need to verify a few things:

• performance isn't negatively impacted on other platforms (single cpu, platforms that don't support sbt, virtualized hardware)
• performance isn't negatively impacted on other backends (leveldb, memory, most of the testing has been on bitcask and memory so far, but signs for leveldb are good).

[evanmcc]

my testing suggest that it's better for leveldb in most if not all
situations. Haven't done much on #1. Maybe I'll just commit the change
for develop right after 2.0 so we have the whole cycle to notice/work on
potential regressions.

On Mon, Apr 21, 2014 at 9:25 AM, Jared Morrow notifications@github.comwrote:

I don't have a strong opinion on this one, though I wonder if the things
mentioned in your first comment have been addressed?

In order to make it a default setting for vm.args, though, we need to
verify a few things:

• performance isn't negatively impacted on other platforms (single
  cpu, platforms that don't support sbt, virtualized hardware)

• performance isn't negatively impacted on other backends (leveldb,
  memory, most of the testing has been on bitcask and memory so far, but
  signs for leveldb are good).

  Reply to this email directly or view it on GitHubhttps://github.com/Validate +stbt nnts as a potential default setting. #391#issuecomment-40948967
  .

[jonmeredith]

l vote for after 2.0. I'll reset this milestone to 2.1