Posts Tagged ‘Eucalyptus

Is someone going to step in and provide commercial support for the Eucalyptus open-source project basically following the Red Hat model which made Linux commercially successful and generated good revenue for Red Hat?

Here’s what I am thinking based on what I saw and heard at the Cloud Computing Expo (see my notes here):

  1. We know that Amazon’s Web Services are rapidly becoming the de-facto industry standard for cloud infrastructure.
  2. The biggest complaints that people have against them are that they are only provided by Amazon, and thus lock you in one vendor and do not support the “private” clouds (on-premise deployment).
  3. Both of these could potentially be addressed by the Eucalyptus project which implements EC2 and (in the next release) S3 APIs and can be used on hardware of your choice.
  4. Eucalyptus team is not planning to use their project commercially and Rich Wolski is skeptical about competing against Amazon.
  5. The project is open source.

In my mind this means that there is significant unsatisfied demand and a technology which someone could use to satisfy it. So my expectation is that we will see someone doing that. If not now then in next year when Eucalyptus is more feature-complete and stable (which is expected in the Spring).

Heck, this could be a very interesting play by Citrix considering that both Amazon and Citrix are using XEN, and Citrix is now trying hard to move beyond terminal access to virtualization and cloud space…

Technorati Tags:
, , , ,


Here are my notes from the third day of SYS-CON’s Cloud Computing Expo (see also my notes from day 1 and day 2):

Peter Nickolov – President & CTO of 3tera – gave a pitch on how their technology (AppLogic) lets customers use cloud computing for high-availability solutions.

In a nutshell, Peter had an instance of SugarCRM which in his demo could fail-over from one datacenter to another. No changes in the code were required everything was set in the configuration of the application and AppLogic: he copied the application (front-end machines, back-end machines, load-balancers, etc.) to another cloud and set MySQL replication between them. Then when one application goes down their load-balancers detect unavailability of the primary site and give the IP address to the secondary one.

Peter said it took a couple of days to set up the demo. Obviously, SugareCRM was a relatively easy target because all the state information is in a single database, so MySQL replication was sufficient to have the application ready for the hot switch. But nevertheless this was a pretty impressive demo of how AppLogic’s building blocks can provide the additional layer of management and datacenter independence you might want to have with your hosters.

Andrew Comas from CORDYS gave a fairly boring general session on their Process Factory product – basically some kind of mesh-up editor for corporate use.

VMware had 2 sessions that day – by Dan Chu (Vice President of Emerging Products and Markets – which at VMware includes everything from overseeing the SMB space, to virtual appliances, to cloud computing) and by Preeti Somal (Vice President R&D Cloud Computing).

Basically, these were a pitch for the upcoming vCloud solution. The basic idea is actually very close to what we get from 3tera and rPath: the cloud is just a set of virtual machines, let’s make them standardized across the datacenters and provide administrators the ability to manage them as a system – and we got a great flexible solution without a hosting vendor lock-in.

vCloud is definitely more of a roadmap rather than a solution you can try:

VMware vCloud Roadmap

Today, they have their existing on-premise Virtual Infrastructure which a lot of us are using in our companies. In addition to that they have over a 100 hosting partners committed to providing this infrastructure in their datacenter – thus providing the flexibility to choose the hosting vendor.

Next year we will start getting into the second phase – so called “vCloud Services”: which basically means that we will get OVF-based way of grouping virtual machines into systems together with associated policies. And we might get a few sample solutions like the “flex capacity” scanario which was demoed during the VMWorld keynote in September.

Finally, they will provide full Virtual Center integration so you can manage your VMs in one console regardless of whether they are deployed in your network or by a hoster (they are calling that Federation) and more advanced architecture capabilities.

It is yet unknown how much will vCloud move beyond just VMs into additional services such as message queuing to VM interaction, storage and so on. They are saying that some of the infrastructure will be provided (for example load balancing) but not everything because they want to stick to creating the common platform which partners will use for the actual solutions.

My bet is that if they want to compete effectively against Microsoft’s Windows Azure and Amazon’s ever increasing set of Web Services they will have to move up the stack and provide more than the basic VM infrastructure. The question is how fast they can move into these new areas and how much the task of keeping all the datacenter partners happy will slow them down.

Their main bets are on application compatibility – just re-use any VMs you have today – and broad hosting partner range. They are also hoping that their vCloud APIs (RESTfull web services) will enable broad ISV ecosystem.

[Download VMware slides]

Erik Carlin from Rackspace‘s cloud computing division – Mosso talked about cloud standardization. This included:

  1. Common taxonomy: Software-as-a-Service (e.g. Salesforce), Platform-as-a-Service (e.g. Google App Engine) and Infrastructure-as-a-Service (e.g. Amazon EC2).
  2. APIs for storage, compute, network, and data. Right now, even when APIs are common across datacenters (e.g. with vCloud and 3tera) you still get locked into application vendor providing you these. Something like Red Hat’s Libvert abstract hypervisor API could help. Ruben from Enomaly pushing that through the Cloud Interoperability Forum.
    Other issues include identity (Erik thinks that OpenID has the biggest potential here while WS-* will probably be used by Microsoft only), and dependencies on particular cloud services.
  3. Pricing Complexity. How do you actually calculate the compute power provided and what is the Standard Processing Calculation unit? Work on virtualization benchmarks by VMware and Intel can help.
  4. Compliance issues: depending on the industry and application you might get to adhere to HIPAA, SAS70, PCI, or Safe Harbor (Rackspace is certified for the latter, European thing for datastorage).

Overall, common standards should provide for interoperability, lock-in avoidance, fail-over scenarios, better tools for all, cloud bursting and multi-cloud applications – which will enable positive network effects and increase the overall market for everyone.

[Download Erik’s slides]

Next we had Rich Wolski presenting his Eucalyptus project – an open-source clone of Amazon’s EC2 and S3.

Rich is absolutely amazing and his sessions are definitely a must-attend. He talked a lot about the architecture of their solution and how people are using it to try/test their EC2 solutions before deploying them with Amazon:

  • They currently have about 80 downloads a day. Download requires no registration so they do not know how exactly they are being used.
  • The biggest installation Rich knows includes 260 nodes.
  • He does not believe that Eucalyptus can be used to compete against Amazon – you still need people, datacenter infrastructure, know how to do machine rollover and so on
  • They currently have 5 engineers on the project and drop monthly releases. At the moment they do not accept external contribution but might start doing that in spring when they stabilize.

[Download Rich’s slides]

Finally we had Gerrit Huizenga – Solutions Architect from IBM and part of their cloud taskforce share his views on cloud computing. I was surprised that he was actually downplaying the role and newness of cloud computing as much as he could but I guess that is part of being from an established corporation with huge established software and consulting business.

[Download Gerrit’s slides]

That is it for my day to day notes. I will also publish my summary notes once I recover from all the recent traveling and catch up on my email.

Technorati Tags:
, , , , , , , , , , ,

Is VMware’s vCloud – a killer for Amazon’s EC2 services? It definitely sounds like they are shooting to get into the same market.

In a nutshell, Amazon allows you to have your (XEN-based) virtual machines run in Amazon’s data center and execute your application. This is great and makes Amazon a leader in that space today but there is a lock-in (as Richard Stallman was saying the other day) – all the APIs are proprietary and your application needs to be designed for EC2 and will only work there.

There is of course the Eucalyptus project seeking to create an on-premise open source version of EC2 but at the moment it is more of an experiment rather than something commercially available for enterprise use.

However, the point is that people do want that kind of portability, and that is what VMware’s vCloud is designed to provide. The vision is that if you have both on-premise virtual infrastructure and external in-the-cloud datacenter infrastructure both built on VMware technology, than you can run the same virtual machines/appliances anywhere you want:

So you are kind of locked to the technology (VMware) but not to a datacenter provider.

At VMworld, the scenario which was demoed had a company start a VM in external datacenter in addition to the ones they had on-premise to offload a part of the peak load. See Michael’s video here.

This is obviously just one of the scenarios. The value of common infrastructure spanning across on-premise and cloud environments goes beyond that and provides the flexibility of using the best resource available at particular time for particular task, and be able to quickly get new resources when needed or reconfigure your virtual datacenter on the fly.

The biggest problem with that, is that for VMware, this is more of a vision thing (despite a line-up of service partners promising to support the platform). It will obviously take time and execution to make it a reality.

Tags: , , , , ,

RSS My company’s main blog

My Recent Tweets



The posts on this blog are provided “as is” with no warranties and confer no rights. The opinions expressed on this site are mine and mine alone, and do not necessarily represent those of my employer Jelastic or anyone else for that matter. All trademarks acknowledged.

© 2008-2012 Dmitry Sotnikov

%d bloggers like this: