Open Stack

On 12/01/2013 02:16 PM, Robert Collins wrote:
> This is for input to the TripleO design prioritisation discussions
> we're having at the moment.
>
> We have been assuming that everyone has almost no variation on
> hypervisors, but specialised hardware configuration for control plane
> / data storage / compute - so typically 3 profiles of machine.
>
> But, we'd like a little data on that.
>
> So - if you're willing, please reply (here or directly to me) with
> answers to the following q's:
>
>   - how many hardware configurations do you have

5 "profiles":

* "high support" nodes (used for database cluster nodes, message queue 
cluster nodes, some other intensive ops stuff. Generally these have 
battery-backed write caches, 128+ GB memory, and a fair bit of attached 
disk storage, in a RAID 10 or RAID5 setup.

* "low support" nodes -- used for less intensive ops activities like 
LDAP slaves (for host access control, not cloud access control), 
graphing, logging, Quantum L3 router nodes, jump hosts, Chef servers, 
etc. Less memory than high support, but varied attached disk storage.

* "controller" nodes -- used for all stateless OpenStack API services. 
32-64 GB memory, little attached disk, 1 processor, varied number of 
cores 6-12.

* "compute worker" -- used for Nova compute workers. Typically 256 or 
144GB memory, 2 processors, 24 cores. 4-6TB direct attached storage, 
typically in a RAID 5 configuration.

* "object storage" -- used for Swift proxy, object and some support 
services.

>   - are they deliberate, or organic (e.g. optimising for work vs your
> vendor couldn't deliver the same config anymore)

LOL. Mix of both? :) We have a physical design / reference architecture 
that our team has put together. It is used for most of our modern 
deployment zones... but money, politics, and agendas can get in the way 
of consistency ;)

For example, we have Dell, IBM, SuperMicro, Quanta (regular and RackGo) 
all in use in our deployment zones. I heard this morning that HP blades 
are going to be added into that mix for some specialized deployments.

And that is just the compute and object storage hardware.

If you also take into account the network gear, it's just as varied. 
Arista, Cisco, Juniper, and Quanta gear is all used in our deployments, 
with lots of models of each vendor.

About the only thing that *is* consistent in our deployments is NAS. We 
use NetApp filers for NFS/iSCSI volume management.

We are 100% heterogeneous, and we have no plans to become homogeneous 
any time soon. At a large company like AT&T, frankly there are just too 
many people involved in both the procurement and 
certification/validation of hardware to ever expect a single-vendor 
environment to be a reality.

>   - if you have multiple hypervisor hardware configurations, do you
> expose that to your users (e.g. using different flavors to let users
> choose which servers to run vms on) - a story for that might be high
> performance computing vs densely packed vms.

We don't vary our hypervisor offering (to date). We offer our tenants 
some scaling improvements, particularly for the Windows tenants, using 
GridCentric's technology, but everything to date has been KVM. No custom 
flavors.

> Thanks!
> -Rob
>