Open Stack

Jordan Pittier said on Mon, Nov 23, 2015 at 11:59:35AM +0100:
> Is the DB the limiting factor of openstack performance ? De we have hard
> evidence of this ? We need numbers before acting otherwise it will be an
> endless discussion.

Properly parallelised workers scale linearly. Persistent systems scale
slower because achieving consistency requires at least NlogN commlinks
between persistence nodes. This generally makes them the slowest part of
any well-engineered system running at the limit.

This is in the abstract though. You're right, we shouldn't get excited
about theoretical gains until we have Clint's numbers. It's quite likely
we're hitting ourselves in the face algorithmically. If constraints are
an issue, we can start removing the binding ones while retaining the
app-level constraints and normal form.

> When I look at the number of race conditions we had/have in OpenStack, it
> seems scary to remove the FK in the DB. FK look like a "guardian" to me and
> we should aim at enforcing more consistency/integrity, not the contrary.

It's always scary leaving the nest. That is the proper reaction! Still
doesn't mean we should stay home forever, just be cautious.

> Also, this is an open source project with contributors with different
> skills and experience (beginners, part time contributor etc.). Maybe this
> is something to also consider.

What are you saying?


Alexis (lxsli)
-- 
Nova developer, Hewlett-Packard Limited.
Registered Office: Cain Road, Bracknell, Berkshire RG12 1HN.
Registered Number: 00690597 England
VAT number: GB 314 1496 79