Have a look at this MSDN blog post to see a real life example of how crucial it is to fail fast.
Sunday 6 September 2009
Tuesday 1 September 2009
I know more or less what Cloud Computing is but until recently I still struggled to figure what it is good for. That’s why I decided to attend CloudCamp at Google’s Sydney Office which is a Cloud Computing event focused on sharing thoughts in a form of open discussions. The presenters were there just to start conversations and finish them soon enough to have a few beers afterwards :). I participated in two sessions(Scaling Applications in the Cloud, Cloud Computing from business perspective) and each of them taught me something different.
The sexiest part of Cloud Computing is its promise of scalability. The funny thing is that most of the people will never need it. If I’m not mistaken StackOverflow handles 1 million of hits a day running on a beefy DB server and two medium size Web servers. Sure, it depends what your app does but in most cases thinking a lot about scaling issues before you actually face them is a waste of time. It’s better to spend that time on making sure your app is successful. This of course doesn’t justify complete lack of design and messy code. It’s a matter of striking the right balance. If you know for fact that your app needs to handle huge load then it makes sense to design for it upfront. But again, if there is one problem that startups are dreaming of it’s the load problem :).
One of the presenters mentioned that hosting of a regular WordPress based site in the Cloud is 7 times more expensive than regular, dedicated hosting. The cloud seems to be good for apps which resource utilization is either low or variable. If it’s low then it means that by hosting it yourself you pay for resources that you don’t take advantage of. If your utilization is high it might not make sense to move to the Cloud because the Cloud Computing providers charge you for resources (you will use the same amount) and additionally for promise of more resources when you need them. If you have to handle spikes Cloud Computing might be the way to go as you don’t want to buy bunch of servers that you use only a couple of weeks a year. In other words the key to successful migration to the Cloud is to know your application capacity.
A few people mentioned that they use the Cloud as their stress testing environments. This actually makes a lot sense because you can quickly provision a lot of boxes, do your testing and then discard them for a very reasonable price. In general if you need to perform some kind of activity that is either temporary or repetitive you might want to consider doing it off your premises.
Another presenter said that the Cloud Computing price should drop in the near future because more and more people are using it and it might become a commodity. Someone compared Amazon pricelist to the price lists of mobile network operators. At the beginning the price list was simple to attract customers. The more people use the service the more complicated the price list gets until the point when a regular user is totally lost and the service provider can make more money off him/her then before. It is an interesting theory and definitively true with regards to at least some mobile network operators. I still can’t figure out why 50 AUD worth CAP converts into 150 AUD value :).
An employee of of a big hardware vendor mentioned that some big Cloud Computing providers are working on a standard that will let people to move their apps from one provider to the other. I suppose they figured out that being locked to a particular provider is not what business is looking for.
From business perspective IT infrastructure is a cost. If the cost can be lowered by moving it to the Cloud that’s great. It’s not like business is mad about Could Computing. If a bunch of monkeys were good enough and cheaper than on premises IT infrastructure they would go for it. IT is a tool. So far the best one and let’s try to keep it that way :).