There is a roundup of QCon by David Norfolk that was forwarded to me by JulesLt that brings up some interesting questions. In the article he mentions what he saw from eBay:

I mostly attended the SOA track and Dan Pritchett, an eBay Technical Fellow, provided some interesting insights here, including the worrying possibility of deadlock between SOA services. Remember “deadly embrace”? And, for instance, “scalability is just a design problem” (it helps if you design for manageability and monitoring, of course) – but what really keeps him awake at night is large scale testing and large scale software and content deployment (if it does go wrong, how on earth do you roll back). Again, it’s a cultural thing, eBay rather makes a fetish of “no single point of failure”. He also pointed out that power consumption and cooling are now a primary constraint to growth at eBay.

Tariq Ahmed and I have chatted about this before, but for large companies like eBay, something like the Apollo desktop application can save them a lot of money. They can offload a lot of CPU intensive activities off to the client. Done on a large scale, this could have implications for the cooling and power problems.

It isn’t limited to Rich Desktop Applications either, any client side technology means less work for the server and more pushed off to the client. If you’re using technologies like LiveCycle Data Services, the server is pushing out information so it doesn’t have to be bothered with clients continually requesting. In large batches, that’s significant savings.

[tags]Rich Internet Applications, Enterprise, Flex, LiveCycle Data Services[/tags]