2 replies to this topic

[Franky]

This is my first post on these forums. Nice to meet you all :)

I am speculating an upcoming project and have never experimented with distributed applications. I need to build an application, that can be deployed on multiple nodes/servers and provide a single interface (ip/domain) to access it from. A perfect example would be facebook ;) . How exactly could one domain name handle infinite requests?! I looked into hadoop, and searched for clustered solutions for basic php/mysql apps but couldn't find anything. I've also read up on consistent hashing but figured it was just a solution for data redundancy. Any input would be greatly appreciated!

[Orjan]

at first, I can say that there isn't only one IP behind facebook and such. They will probably have an dns pool which is the first part, it more or less randomizes which of the to the host assigned IPs that user will see the site as. This is manageable just by adding several a-records for the same host name, if I remember right. MySQL in their enterprise solution has cluster solutions, so that is the next step to set up. that's a small part of it at least. but there are so much more.

[John]

Facebook also has two layers of data abstraction. They cache the majority of their data, and when the cache is out of date, it is updated from the persistant data store.