Comments on: Thoughts on the Economics of the AWS Cloud http://blog.anthonychaves.net/2009/12/12/thoughts-on-the-economics-of-the-aws-cloud/ Life is software and jujitsu Mon, 01 Feb 2010 17:22:12 +0000 http://wordpress.org/?v=2.9.1 hourly 1 By: Chris Adkin http://blog.anthonychaves.net/2009/12/12/thoughts-on-the-economics-of-the-aws-cloud/comment-page-1/#comment-9533 Chris Adkin Wed, 16 Dec 2009 22:16:24 +0000 http://blog.anthonychaves.net/?p=153#comment-9533 Hi Anthony, Someone from Packt publishing contacted me several weeks back with regard to writing a review about your WebSphere Extreme Scale book. Thus far, based on what I have read, the material is well constructed, practical and very accessible in terms of starting off from the basics and providing code samples throughout. As Billy Newport has reviewed this, I would presume that it is rock solid in terms of technical accuracy also. I would guess that IBM will evolve eXtreme Scale to allow things such as the annotation based injection of object grids etc into the application code. I would also be interested to know if the grid configuration xml can be boot strapped by being embedded in Tomcat web.xml files, as per the popular method of boot strapping Spring application contexts. On the subject of storage and the cloud, Martin Fowler mentioned at a ThoughtWorks get together not so long ago, that compute intensive applications scale well in the 'Cloud' but I/O intensive apps do not. I would go along with this. By their nature clouds should be opaque, therefore you have no idea if the underlying storage is raid 5, good for read intensive workloads, but not so good for write intensive workloads. Also, how does a cloud segregate random OLTP I/O from the sequential style I/O ubiquitous with data warehouses and business intelligence applications. Some storage arrays suspend all I/O until the cache is flushed to disk, therefore your application is going to suffer if you are sharing an underlying array with someone elses app(s) that likes to fill the cache. The classic mistake when building systems is to build unbalanced systems whereby there are insufficient host bus adapters, pipes, storage processors and spindles to feed the CPUs. I would like to know if Amazon contractually guarantee sustained I/Os per second and what tooling or APIs they provide to monitor this. I would also be interested in the levels and quality of support that Amazon provide to customers using EC2. This may be a potatoes to apples comparison, but I can say from experience that dealing with Amazon over a disputed market place transaction was about as easy and straight forward as trying to contact the dead !!!. At my last company due to contractual requirements, the ops guys had to perform disaster recovery exercises at dr sites, how does this work with the cloud ?, the case of sidekick data loss springs to mind in particular. Refer to this story about “When the cloud fails”: http://gigaom.com/2009/10/10/when-cloud-fails-t-mobile-microsoft-lose-sidekick-customer-data/. Regards, Chris Hi Anthony,

Someone from Packt publishing contacted me several weeks back with regard to writing a review about your WebSphere Extreme Scale book. Thus far, based on what I have read, the material is well constructed, practical and very accessible in terms of starting off from the basics and providing code samples throughout. As Billy Newport has reviewed this, I would presume that it is rock solid in terms of technical accuracy also.

I would guess that IBM will evolve eXtreme Scale to allow things such as the annotation based injection of object grids etc into the application code. I would also be interested to know if the grid configuration xml can be boot strapped by being embedded in Tomcat web.xml files, as per the popular method of boot strapping Spring application contexts.

On the subject of storage and the cloud, Martin Fowler mentioned at a ThoughtWorks get together not so long ago, that compute intensive applications scale well in the ‘Cloud’ but I/O intensive apps do not. I would go along with this. By their nature clouds should be opaque, therefore you have no idea if the underlying storage is raid 5, good for read intensive workloads, but not so good for write intensive workloads. Also, how does a cloud segregate random OLTP I/O from the sequential style I/O ubiquitous with data warehouses and business intelligence applications. Some storage arrays suspend all I/O until the cache is flushed to disk, therefore your application is going to suffer if you are sharing an underlying array with someone elses app(s) that likes to fill the cache. The classic mistake when building systems is to build unbalanced systems whereby there are insufficient host bus adapters, pipes, storage processors and spindles to feed the CPUs. I would like to know if Amazon contractually guarantee sustained I/Os per second and what tooling or APIs they provide to monitor this. I would also be interested in the levels and quality of support that Amazon provide to customers using EC2. This may be a potatoes to apples comparison, but I can say from experience that dealing with Amazon over a disputed market place transaction was about as easy and straight forward as trying to contact the dead !!!. At my last company due to contractual requirements, the ops guys had to perform disaster recovery exercises at dr sites, how does this work with the cloud ?, the case of sidekick data loss springs to mind in particular. Refer to this story about “When the cloud fails”: http://gigaom.com/2009/10/10/when-cloud-fails-t-mobile-microsoft-lose-sidekick-customer-data/.

Regards,

Chris

]]>