robert.collins at canonical.com
Thu Jun 2 20:07:40 UTC 2011
On Fri, Jun 3, 2011 at 7:48 AM, Dustin Kirkland <kirkland at ubuntu.com> wrote:
> On Thu, Jun 2, 2011 at 8:02 AM, Tom Haddon <tom.haddon at canonical.com> wrote:
>> == Things we don't like about Ensemble ==
>> - Ensemble seems to currently require a cloud infrastructure (EC2/S3
>> specifically) to run. Are there plans in the future to allow Ensemble to
>> run on bare metal? Our usage of EC2 has been limited for a number of
>> reasons, including cost and performance. If the plan was to only ever
>> have Ensemble work on EC2, that'd make it hard to adopt it for our
> "Bare metal" is precisely the intersection point of the Ubuntu
> Orchestra project (http://pad.lv/p/orchestra), and Ensemble.
> We're rapidly evolving Ubuntu Orchestra in the Oneiric archive right
> now (and will be introducing it properly and publicly very soon). But
> for the sake of this discussion, Orchestra is a mechanism by which
> bare metal systems are rapidly and automatically provisioned (much
> like ec2-run-instances does for the cloud). The Orchestra-managed
> bare metal system is installed, rebooted, and taken to the point at
> which it runs cloud-init, and "discovers" its identity (perhaps as an
> Ensemble Agent).
> Ensemble's interaction with Orchestra is still in the discussion
> phase, but rest assured that the Ensemble Team and the Server Team are
> discussing these touch points and will take this bit of feedback from
> your team as an essential implementation goal.
There is a nuance here that perhaps is missed: one of the performance
things Tom talks about is the time-to-change.
Launchpad runs with 4 fairly beefy appservers (roughly equivalent to
m2.4xlarge but with less memory). This is sized to handle all our
traffic without *significant* degradation with only 3 of the servers.
If a deploy takes out each server serially for 3 minutes, we can feel
fairly happy that we're not going to have one of the as yet untouched
servers melt itself before the server thats being replaced is out.
Adding in *either* a fresh install of Ubuntu *or* building a chroot
*or* copying over an AMI and booting a virtual machine seems like it
will have to add that time to do that on top of the time for the
deploy of that machine. One of the cloud answers here is 'turn on more
capacity before you turn off the old servers' - but we're hosted
internally, so that capacity means there has to be a 5th machine
available (and this still applies with lxc, it just changes from
'machine' to 'memory and some CPU' - which all the appserver nodes
So at least AIUI what Tom is comparing with isn't VM's *specifically*
but 'here is my OS, its a stable platform, and without nuking down to
fresh-installed state, please get a new version of Launchpad running
More information about the Ensemble