Q&A on OpenStack + VMware

We recently held a webinar on “Architecting OpenStack in your enterprise” together with Gigaom, which produced quite a lot of buzz. Our many attendees had a lot of questions which we weren’t able to answer due to time restrictions. We promised to follow up, so here is a list of these questions and answers. We hope you find them informative!

1. How do you monitor network traffic from VMs moving dynamically between multiple hosts without losing visibility of moving VMs?
Assuming the instance id remains constant between migrations, you can query the instance id using OpenStack Ceilometer, the OpenStack metering service, linked to OpenStack Neutron.

2. Are there any plans to provide warm-upgrades for OpenStack? As you know, even cold upgrades didn’t work till now as it should and its a real pain point.
At the OpenStack Summit in San Diego in 2012 Mark Shuttleworth demonstrated live on stage upgrading from a live Essex based cloud to Folsom in 3 minutes using Juju. Warm upgrades are certainly possible with the right tooling although clouds with sophisticated virtual networking (Neutron) setups are still tricky. See: http://www.youtube.com/watch?v=bcwqvAFBQVg

3. Could you please provide any insights to TripleO project, does TripleO going to become some kind of default Distro for OpenStack? and could someone use TripleO to deploy OpenStack on baremetal?
TripleO is not an OpenStack distribution but a project to create a common set of tools for deploying and maintaining OpenStack. TripleO can be used to deploy OpenStack onto baremetal using another OpenStack related project called Ironic. Both TripleO and Ironic are relatively new solutions to requirements that Canonical has been addressing with Juju & MAAS for over a year.

4. Does OpenStack support IPv6?
Yes it does. Havana has limited support for IPv6 in Neutron. Advanced support is one of the key objectives of IceHouse.

5. Can you show the results of the last survey – is OpenStack ready for enterprise?
Yes, we are currently working on analysing the results and will post a summary on insights.myasnchisdf.eu.org in due course, so stay tuned!

6. If we deploy openstack using packstack, so can we configure individual OpenStack component or add OpenStack component later?
It would be best to check with the Packstack development team regarding their intentions with packstack https://wiki.openstack.org/wiki/Packstack.
On Ubuntu, Juju and MAAS, are two mature OpenStack deployment technologies that provide this capability today.

7. How can we plan to deploy OpenStack on large scale enterprise IT?
Canonical is a in privileged position because Ubuntu is the reference platform for OpenStack deployment and development. We have helped service providers, telcos, and large financial institutions in every phase of their OpenStack deployment.
The projects that moved the fastest had representation and sign-off from every stakeholder within the IT (network, storage, compliance) and IT end users (apps dev, test team, etc.) to investigate and pilot the environment. Canonical professional services can assist with the pilot, address concerns and provide clarifications, and help teams jump the chasm from pilot to production.

8. Is juju intelligent enough to understand if its tier 1 data vs tier 2 data and move data b/w public and private cloud?
Juju itself does not distinguish between data tiers. Nor does it move data around between clouds — but Juju charms can do whatever they want, and it’s possible to connect proxy charms to remote services and use charm logic to automate such things.

9. In term of a) application point of view b) IaaS point of view. what is the difference between (vmware , hyperv) vs OpenStack?
That’s a broad question indeed, and depending on your use case and the VMware/hyperv technologies involved, somewhat difficult to answer easily. We address some of the differences between VMware and OpenStack in our upcoming webinar which you can register for here.

10. Any tool to make sure dependencies between different projects and versions are compatible – I don’t want to break the system with ”non compatible” installation or upgrade by one component ? Any configuration mgmt tools integrated to OpenStack – Puppet etc.?
Interoperability between OpenStack components is a concern for OpenStack users and the challenge grows as the number of technologies in the OpenStack ecosystem expands. Many configuration management tools work with OpenStack, including Puppet, Chef and Juju and each can help deploy and manage an OpenStack environment although true interoperability or integration requires comprehensive testing of all the various permutations of OpenStack technologies. To address this Canonical launched the OpenStack Interoperability Lab (OIL) which tests OpenStack in different configurations using hardware and software from across the OpenStack ecosystem. OIL now tests over 3000 permutations per week to be able to quickly identify non compatible components. Juju manages this environment and is the tool we’d recommend you use to manage deploying compatible components

11. How much better is OpenStack Compute, networking and storage (in performance, in costs) compared to competitors ? Any measurements or benchmarks really available?
This is very tricky to answer as everyone can find different results based on their needs and usage and no truly independent benchmarking between competitive solutions exists as far as we’re aware. In terms of momentum though, as an open source cloud solution OpenStack is the undoubted leader with the support of many vendors and huge growth in adoption.

12. What is the monitoring component of OpenStack (monitoring all layers, HW, SW, connectivity, storage, security, end to end application performance, databases etc )?
There is no single tool for managing every layer of the OpenStack environment. The reality is that most users have combinations of tools that suit their requirements or are provided as part of their OpenStack distribution of choice. We see many users using standard tools like Nagios for monitoring of services and the OpenStack documentation has good examples of how to set this up. For other layers in the stack, people use hardware specific tools for hardware monitoring and OS specific tools for monitoring of individual server status. With Ubuntu we recommend Landscape to manage and monitor your Ubuntu OpenStack cloud.

13. If a customer has virtualized on vmware (who are the large virt. vendor in the market today), why should they build a cloud on OpenStack rather than vcloud?
Many organisation are looking to retain the use of their VMware virtualisation estate while wanting to reap some of the benefits of an open cloud platform like OpenStack.
Things like open APIs, the fast moving innovation, open SDN solutions and many other features are driving people to consider new options in their datacentres.

14. Why should I build an OpenStack private cloud rather than just use readymade amazon AWS?
There are two main reasons that an organisation may choose to switch from public cloud providers to private cloud, including cost and privacy/data protection. With respect to cost, many organisations start out small in AWS, prototyping and testing things out, not necessarily providing production services.
Developers/admins very quickly get used to the ability to rapidly deploy new instances and services which causes costs to ramp up very quickly, often almost organically production services end up getting “accidentally” deployed there. Once the usage hits a certain tipping point there’s a decision to make between continuing to allow the costs to increase externally, or bring the services into the organisation’s own datacentres, it’s purely a case of comparing the cost metrics to ensure that it makes sense to do so. If a decision is made to bring the services in-house then the organisation will need a comparable IaaS platform, of which OpenStack is an obvious solution.
Regarding data privacy, many organisations have regulatory, compliance or security regulations that state they must maintain internal control of sensitive data which may be related to customers, R&D or finance. This data may not be suitable for hosting on public cloud environments.

15. Do you foresee a trend migrating from cloudstack to OpenStack? Any foreseeable timelines?
We have seen a couple of examples of customers moving from CloudStack to OpenStack, but in general, most customers are using OpenStack as their first experience with Open Source cloud infrastructure.

16. Would virtualization have an advantage over bare metal when it comes to hosted hadoop as a service in a private cloud enviro?
Virtualization has an extra overhead not found when using bare-metal. This overhead comes from creating and managing the virtualized operating system. As such, from a pure, outright performance perspective, hadoop on bare-metal is currently going to be the optimal option. However, from a cost and ROI point of view, many customers see the flexibility offered by using virtualised infrastructure outweighing the performance advantages.

17. I think a big question for enterprise is i/o performance for the guests. In particular, what is the current best practice for enterprise? OpenStack block level storage, something zfs based, or a more traditional solution (SAN or NAS)- what are people seeing out in the real world? What is and isn’t fully baked?
There are two different levels to cover:
The physical storage infrastructure: Here the best practice is to follow VMware vSphere recommendations of your storage vendor. Shared storage in any enterprise-level setup is highly recommended. It needs to be capable of handling a large number of IOPS that will be produced by the OpenStack instances OS and by vSphere when creating, deleting or doing any operation in the vSphere datastores (VMFS operations)
The OpenStack infrastructure: As the instances sit on top of vSphere, which is already a very mature technology, the recommendation for OpenStack volume/block storage is to use Cinder with the VMware VMDK driver which will take advantage of all the underlying optimizations and unique features transparently, such as VAAI or Thin Provisioning. The same applies for the ephemeral storage which will natively use vSphere storage.
It is technically possible to set up a separate storage environment for Cinder but it introduces complexity to the architecture and it is recommended to study carefully the potential benefits vs complexity.

18. When will we start seeing comparisons between OpenStack and other Cloud management tools such as from VMware or other vendors?
Gartner and others have offered their views on the comparative maturity of solutions but so far there have not been any in depth, feature by feature comparisons made publicly available. We expect these to come soon as OpenStack gains in maturity and popularity although the really interesting questions are how these different environments can be connected and integrated as rarely is a customer making a straight either/or comparison.

19. Do I need to use Vcenter if I use OpenStack for provisioning on ESX?
It depends. You will need it if you want the vCenter management capabilities to extend to OpenStack for a cluster of vSphere hosts. The Havana version of the vSphere OpenStack Virtual Appliance includes the new vCenter Server plug-in for OpenStack frameworks. The plug-in provides vSphere administrators the ability to identify OpenStack instances and some of their respective properties from the vCenter Server. The plugin is jointly supported and certified by Canonical and VMware.