[Openstack-operators] Implementing Openstack, some questions..

[Gerrit Tamboer]

Hello fellow Openstack addicts! 

First, a short introduction; 

I am a
Unix/Linux engineer working for a Hosting company.
We are planning on
building a cloud that is based on Cloudstack for hosting online
applications and other 'Virtual Machine' related software. 

After some
testing we decided to go for the following setup to start off with:


==================== 

2 x Nova Compute servers 

1 x Loadbalancer for
Glance nodes and SwiftProxy nodes (LVS-TUN + Keepalived) 

2 x Glance
servers (Loadbalanced / With a MySQL back-end, master-master replication)


2 x Swift Proxy servers (Loadbalanced / Sharing the Memcached server) 

3
x Swift Storage servers (With 8 TB of storage, devided in 3 zones (1 server
per zone) 3 replications per device (which means 8TB of total space))


==================== 

We are pretty confident about this setup, but we
have some question marks what we would like get answered before we order
the actual hardware :). 

1. In our cluster we have the Swift Proxy servers
load balanced, they share their memcached servers with eachother. What I
know is that auth tokens and account information (etc.) is put in memory
for faster fetching. But does that mean that the auth tokens that are
generated on the first server is also saved in the memcached of the second
server? Or would this setup cause problems when one of the servers go down?
(De loadbalancers are smart enough to skip the server that is not
responding, that should not be an issue) 

2. We would like to know if we
could combine the Swift Proxy nodes with the Glance nodes. In our setup
that are some steps from Nova to the Storage nodes, we would prefer if one
step would be taken away by combining 2 services (in this case Swift Proxy
and Glance). Is this possible? Or would that become a major bottleneck?


3. In the Glance Api configuration there is an option
'image_cache_enabled'. What does this actually do? How does it impact
performance? 

4. We have some trouble deciding what to do with hardware.
For the Swift Storage nodes we are focussing on total amount of storage.
The Proxy nodes need allot of CPU, same as for the Glance nodes. The Nova
servers need allot of ram and CPU because they will host the actual Virtual
Machines. Is this correct? Or am I missing something important? We don't
want hardware to be a bottleneck! :) 

Thanks in advance for all of your
answers! :) 

Regards, 

Gerrit 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20110927/9d284c69/attachment.html>