[openstack-dev] [kolla][osic][bifrost] OSIC cluster status
Steven Dake (stdake)
stdake at cisco.com
Sat Aug 6 00:48:24 UTC 2016
Thanks for the video, those are great - if not a little hard for the
casual observer to follow.
That said, benchmark results of scale testing can be a dangerous game to
play, so lets not play games about it.
I'd like to see a full characterization of the hardware gear and topology
under test so others can duplicate the results if they so desire. Lets
get this stuff in the repo in the doc directory asap - starting now (or
Monday, since its the weekend and our team needs some well earned R&R).
I'll login to the OSIC cluster this weekend and get the documentation
started with what I think would be helpful information for folks
evaluating scale testing benchmarks. Since I wasn't actively involved
with the gear setup, I will be requiring help (in a starred set of
reviews) to correctly identify the hardware under test, how the networks
are setup, how the filesystems are setup, how the raid is setup, how
docker is configured, the exact make and model of the CPU, chipset, brand
and model of nvme SSDs, and memory bandwidth (I assume its quad channel
memory, but more details are needed). I'd like to objectively benchmark
performance of /var/lib/docker, the root filesystem, and the memory
bandwidth of the using respected third party benchmarking tools.
The results in this thread are just preliminary to share with the
community, but please wait for the final documentation in our repository
to be completed in about 3 weeks to pass judgement on the scale
One thing that is clear from the video Michal produced is indeed Kolla can
scale to 4 controller nodes and 101 compute nodes (IIRC I believe on IRC
he indicated this was the system configuration used during our dead
chicken testing) with no problems other then human error (which Sean
Mooney is rapidly working to eliminate via his team's bare metal
deployment work with BiFrost thanks in no small part to the contributions
of the BiFrost and Ironic community). The deployment is very fast (20
minutes) on very fast hardware (the OSIC kit is super fantastically fast).
We also intend to measure day 2 scenarios (such as upgrade and
reconfigure) with a variety of configuration options (including ceph as a
One thing that wasn't mentioned in these threads is some basic testing of
the OpenStack cluster was done to validate it was functional in the
control and data planes of the system.
These results are real with video evidence but as I'e stated the
characterization is incomplete.
I'd like to thank OSIC for facilitating this community effort to help
gather some real-wrold benchmarks of how well atleast Kolla performs in
deployment and operational functionality.
If folks have other interests in specific data please weigh in. Once our
loan of the hardware is up, it may be some time before we have another
shot at using it.
We are far from finished with using the OSIC cluster - we have a tough
slog ahead finishing our scale testing benchmarking results - thanks to
the Kolla community for sticking with it and doing their best to produce
objective results for consumption by a variety of individuals.
I'm sure there is something I missed in the need to characterize the
systems under test, so if I missed something above, please point it out so
we can fix it up front.
Thanks and fantastic work!
On 8/5/16, 11:53 AM, "Michał Jastrzębski" <inc007 at gmail.com> wrote:
>And we finished our first deployment
>We had some hurdles due to misconfiguration, you can see it in along
>with a fix. After these fixes and cleanups performed (we don't want to
>affect resulting time now do we?), we deployed functional openstack
>successfully within 20min:) More videos and tests to come!
>On 5 August 2016 at 11:48, Paul Bourke <paul.bourke at oracle.com> wrote:
>> Hi Kolla,
>> Thought it will be helpful to send a status mail once we hit
>> the osic cluster work, so people can keep up to speed without having to
>> trawl IRC.
>> Reference: https://etherpad.openstack.org/p/kolla-N-midcycle-osic
>> Work began on the cluster Wed Aug 3rd, item 1) from the etherpad is now
>> complete. The 131 bare metal nodes have been provisioned with Ubuntu
>> networking is configured, and all Kolla prechecks are passing.
>> The default set of images (--profile default) have been built and
>> a registry running on the deployment node, the build taking a very
>> OpenStack Development Mailing List (not for usage questions)
>>OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
>OpenStack Development Mailing List (not for usage questions)
>Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
More information about the OpenStack-dev