3. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
What is a Private Cloud?
Elastic or Traditional Virtualization (hint: Pick One!)
Multi-tenant (Multi-Application)
Size
Private API endpoints
Limited inbound connectivity
Customized for specific workloads
4. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Build with the End in Mind
# of instances on this physical host?
Flavor: m1.tiny
Ram: 512MiB
vCPU: 1
Disk: 0 (size of image)
Total instances:
?
48
vCPU Ram Disk
5. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Build with the End in Mind
# of instances on this physical host?
Flavor: m1.medium
Ram: 4096MiB
vCPU: 2
Disk: 10GiB + 40GiB
Total instances:
24
vCPU Ram Disk
6. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Build with the End in Mind
# of instances on this physical host?
Flavor: ram heavy
Ram: 64GiB
vCPU: 8
Disk: 100GiB
Total instances:
2
vCPU Ram Disk
7. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Capacity Thoughts
Don’t allow disk size of 0
Public Cloud: Limit flavors to image size
Private Cloud: Can add flavors for specific
application profiles
Network Utilization
8. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Capacity Thoughts
(resource / resource in smallest flavor) = maximum # of
instances per machine
Fixed Network - Double or quadruple number of expected
instances
I can always add more machines....
But you can’t change or modify the fixed
network once instances are running
9. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Build with the End in Mind
2 networks (3 w/ floating IPs)
1) Host Network
2) Fixed Network
3) Floating network
Easy to add physical nodes and/or
networks
Don’t try to change the fixed
network once in production
Easy to add additional floating
networks
nova-network
10. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Images and Storage
Images
Building images talk tomorrow @1:50 in room C123
Format: raw qcow2 vdi ...
Container: bare ami bundle ...
Cloud tech: cloud-init none
Partitioning: fixed dynamic
Drivers: virtio vhost_net xen ...
12. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Images and Storage
Glance Performance
Network Throughput: 1Gbps = 125MBps (expect ~112MB/s)
RAID 5 - large sequential read/writes
Disk Bandwidth > IOPS
Improve cache - Reduce # of images
Image Size Not Cached Cached
1.4GB 20secs 1sec
16.4GB 2min 21secs 1sec
*times from “creating
image” to “qemu-img
create”
13. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Images and Storage
Storage
4 focus points: Glance, Compute (local), Cinder, Swift
Glance - Space, Sequential read/write
Compute - random IO
Cinder - Performance, Density
Swift - JBOD, density
14. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Architecture Examples and Thoughts
1 - 20 physical servers
Single controller
Single API
Single network (1Gbps or 2Gbps)
15. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Architecture Examples and Thoughts
20-100 servers
HA controllers and load balanced APIs
Swift/CloudFiles for Glance
Availability Zones
Consider “frontend” and “backend”
networks
Compute Node metrics - Dedicated
Machine(s)
16. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Performance Considerations and Bottlenecks
IO
Random IO
Reduce IO per instance
Block Storage!
Review Hypervisor best practices
17. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Lessons Learned
Lessons learned
Floating IPs MUST be associated with the “public_interface”
Each piece of OpenStack has its own architecture
Folsom is stable
Migration (live, block) works but scenarios exist where it
doesn’t
OpenStack is still changing often, keep up to date with
current state of the projects
Don’t do Heterogeneous Nodes
18. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
Operational Updates
nova hypervisor* calls
Image types in Glance
policy.json
Coming in Grizzly:
Cells
Quantum
Better LDAP/AD support
19. RACKSPACE® HOSTING | WWW.RACKSPACE.COM
But....
But this is a design summit also
Open to
discussions/thoughts/questions
Goal: To give ideas on how to build private clouds powered by OpenStack software This talk is to serve as a guide not and end all be all. Ultimately you’ll have to find the right solution for your company. Ask to hold questions till the end
Plan for these talks: We’ve been using and deploying Folsom Private Clouds for 6 months, next Summit, we’ll give a Grizzly Talk Grizzly is quite a different ...”beast”... I’ll focus on some of the Grizzly features at the end
Are you looking for a cheaper VMware or a truly elastic private cloud? OpenStack Can do both but it certainly isn’t a drop in replacement for VMware Private Clouds generally not multi-tenant where tenants are different companies. However they will most likely be Multi-application which is in a sense multi-tenant since different applications require different resources. Size for Talk: Less than 100 nodes Inbound Connectivity: Floating IPs or all instances get public IPs
Problem: don’t know how many virtual machines due to flavors CPU/RAM ratio way off, disk most likely under utilized
Problem: don’t know how many virtual machines due to flavors CPU/RAM closer
Problem: don’t know how many virtual machines due to flavors CPU/RAM ratio way off
You’re the service provider, you must plan for capacity. Similar issues as traditional virt but now users can spin up on demand without going through an ops team public model: limit flavors , can pre plan capacity easily private model: closely watch capacity trends since you can’t plan as well Don’t forget about network utilization
pre-quantum you cannot add/change you fixed network -- Some ability to add different networks but users MUST pass which network they want to use at instance boot time fixed range defines maximum number of IPs per “cluster”
Fixed network is the one thing you need to get right. tell why!! Assigning a floating IP completely changes the way inbound and outbound connectivity happens. I’m giving an entire talk about nova-network tomorrow at 11:50am
format - ISO can be used to create initial image with VNC with Raw - use fixed partitioning qcow2 - gain space saving benefits without losing too much performance over raw
Explain what glance is It’s hard to realistically guess how many images or size of images you’ll be using. Expect Users to snapshot It’s simpler to standardize on base images and use automation tools to configure services within the instances.
SIngle Controller - MySQL, rabbitmq-server, keystone, glance registry/api, nova-scheduler, nova-os-api-compute, nova-cert, nova-vncproxy, horizon Network utilization will be different. Understand your current usage and plan accordingly. If worried about nic saturation, break out your nova services (glance, nova services) to a separate network, Aggregate Bonds
Convey why to consider swift 100 nodes, 2000 instances (20 nodes per), any of them could be snapshotting. Will be a bottleneck frontend backend networks frontend for external connectivity backend for instance to instance, instance to non-openstack server (dedicated DB,
If you’re not using another system (Cinder, SAN, NetApp. etc) for additional storage, IO will need to a top consideration
As I mentioned when starting, every major slide on here could be an entire talk and most of these slides are “lessons learned”
nova calls for stats and instances on a hypervisor new image types, includes ISO policy.json allows you to define which role is allowed to make a specific API call Cells - Mainly used for large clouds (500+ nodes) Quantum - default network manager working LDAP/AD integration expected