DAY TWO

Day 2 started with a keynote by Amazon’s CTO - Dr. Werner Vogels. Werner is an amazing speaker, and his enthusiasm really gave a kick-start to the whole day. I would summarize his presentation as: “Yep, we invented cloud computing and we own the space.”

Werner talked about the history of Amazon’s Web Services with the basic idea of that being in line with their traditional platform approach they have been using in the commerce for ages, and also with a lot of IP, experience and investment that went into providing these massively scalable services as a cloud platform.

He gave multiple examples of start-ups such as Animoto and Mogulus which would not be possible without Amazon (because of huge upfront capital costs they would have to bare), companies able of cost-effectively accommodating pick demand (Indicars for their 3 races a year), and application vendors using EC2 for their online versions (Mathematica and MathLab) or demo environments (Splunk).

One interesting tidbit was that recently traffic generated by Amazon Web Services got well ahead of the traffic from Amazon’s store properties (including media sold by them online and so on), and the way the former is growing the latter will be virtually unnoticeable on the combined graph very soon.

He also pointed at how they are working on making the services more enterprise-friendly by providing:

• Way to troubleshoot and eliminate failures,
• Control over data/services geographical location,
• Control over performance (with their content delivery platform - Cloudfront),
• Availability (and adapted licensing) of all major OS and database platforms from Red Hat, Sun, Oracle, and Microsoft,
• Ecosystem of ISVs and system integrators (such as Capgemini).
• Efforts to ensure security (one medical app running on Amazon recently got HIPAA certified.)

David Young from Joyent talked about how they are making the cloud more open and user-centric. In the next 18-24 months Joyent plans to move from traditional and rather meaningless for cloud computing Service Level Agreements (SLA) to what he called Application Performance Level Agreement (like: “99% of my customers are getting 3 second responses 99% of the time”).

They are also working on providing frameworks to ensure:

• Automatic scaling of applications they host (today this has to be developed by application creators themselves or they have to lose control and stick to Google App Engine). Basically, new stacks are needed for dynamic scaling, and they will provide something like Google App engine but without Python and Big Table requirements.
• Ability to move between EC2 and other infrastructure providers.

He also mentioned that Twitter has licensed their technology to run the service in their own private cloud.

After that David Bemstein from Cisco gave his pitch of why networks still matter. He started from the premise of their customers keeping asking them about the right network infrastructure for private and public clouds, but then basically switched to Cisco’s efforts in the virtualization space like their Unified Fabric: lossless (datacenter) Ethernet and virtual layer of simulated fiber-channel on top on it.

More or less the only reference to the cloud after that were his words that Amazon does not provide secure network separation, network within Amazon is basically flat and other vendors could use Cisco technology to differentiate.

[Download David's slides]

Serguei Beloussov from Parallels gave an overview of all the markets in which they compete (obviously everyone is familiar with their desktop virtualization for Mac) and also talked about the clouds. Basically they want to be the provider of choice for the automation for clouds: provisioning, workflow, chargeback, delegation to partners and users, and so on.

They are already fairly strong in the service providers space. For private clouds, they think that Microsoft will not be able to kill VMware (”Microsoft is not what it used to be, and VMware is not Netscape”) - so there will be competition which is good for them because they can have the heterogeneous automation game.

When asked about virtual appliances, Serguei basically said that they don’t have resources to be in that market today but they plan to get there within the next couple of years.

[Download Serguei's slides]

Thorsten von Eicken from Rightscale talked about his vision of the way the cloud should be: no upfront costs, automated scaling (so pay for the average use not the peak), easy to use for dev and test (again with automation to give proper environment set up), easy to use for batch jobs - basically cloud computing wrapped into a layer providing full automation for provisioning and scalability.

RightScale provides some level of automation today: they do monitoring and have automated system which can start and stop new instances of your machines in the cloud as required. However, a lot of scalability needs to be baked into the application itself. He gave an example of Animoto which had to scale 10 times in 3.5 days in April 2008 when their widget became popular on Facebook, and how they had to fight various issues (like log files filling up the disks) as they were scaling up the service.

Thorsten also talked about how they want to make it easier to move (or failover) between providers. There can be a lot of small details in each infrastructure making the switch hard like: server persistence mechanisms, IP address mapping, load balancing, disk volume reassignment and so on. And obviously proprietary platform-specific APIs for storage, queuing and so one add to the problem. RightScale does not have a comprehensive solution for that now, but some of the issues can already be addressed by their machine templates.

RightScale already supports Amazon’s EC2, Rackspace, Flexiscale, and GoGrid rigth from their web UI. They are also very agile - for example, they have already added UI for Amazon’s CloudFront released just a few days ago.

Next we had a panel discussion with Geoff Brown, Ken North, Stefanos Damianakis, and Eric Samson.

They were talking on how there are some low hanging fruit which will go to the SaaS/Cloud world first like Email, disaster recovery, log storage, archiving and so on. Obviously Larry Ellison saying he is not getting the cloud came up. However, as panelists noted Oracle has had hosted versions of all their products for a long time (and has the offering in Amazon’s EC2 world) so he is probably simply looking for right business models to come.

They also talked on how mobile access can make SaaS model predominant, and in general how parts of systems may or may not be able to move into the cloud.

Finally, we had Javier Soltero who is the CEO at Hyperic fighting the myths of operations-free clouds.

His basic idea was that cloud is just a new platform to run applications. Which means that tools to manage them and provide monitoring, performance management, patch management, backup and recovery, budgeting and so one - are still required. In fact most likely new tools are required because the old ones do not support this new environment.

Folks at Hyperic has adapted their existing (open-source) web apps monitoring solution (HQ 4.0 launched last week) for Amazon: application is packaged as an Amazon machine, they are using Amazon’s storage, and their small agents to be placed on the actual application machines can communicate with the service. They also work in mixed deployments when part of the application is in EC2 and the other half - on premise.

They have also launched an interesting cloud availability service at www.cloudstatus.com to measure performance of various cloud engines. They currently do so by firing up and stopping Amazon machine instances but in the future plan to start collecting anonymous data their agents in customer applications.

This was a busy day!