As someone pointed out in the class, I am also curious about how Amazon deal with some holidays like Black Friday when most e-commerce websites including Amazon have a burst of visit? Does this mean Amazon has a much higher amount of over-provisioning servers than actually needed, which seems to be a waste of resource in most cases?
grose
Fun fact: Japan once crashed Twitter by all tweeting about New Year's at the exact same time
@BigFish - Yes, I remember reading somewhere they were at like 25% capacity when not during Christmas. It made for really good DDOS protection. I think they rent that out via their cloud platforms (AWS) now though.
afzhang
@BigFish: Yep! Kavyon mentioned that AWS came about out of trying to recoup some of their server costs during non-peak seasons. The cost of using AWS varies as Amazon's own demand on their servers varies (i.e. you pay a premium if you try to use AWS during the middle of the holiday shopping season).
BigFish
@afzhang Thanks for answering. But my question is that if AWS renting out server during non-peak seasons, what should they do during peak seasons when there are a large demand of servers by users and Amazon itself? Does this imply that Amazon has to keep far more servers than needed during non-peak seasons even though they are renting out some of them.
rbcarlso
@BigFish I don't know the exact answer to this, but I imagine that this is a case where market forces just make it work out. They should bump up the prices for computation power during peak periods to the point where services that DON'T see spikes in usage during those periods temporarily cut back to lower priority services. It's not ideal, and it's a bum deal for those who don't see the extra cost as worth the guaranteed computation power, but it should mean that the resources are being used at high efficiency year round. Recall that you rarely purchase physical processors from AWS; rather you purchase different degrees of performance and reliability.
More demand -> higher cost. Less demand -> lower cost. Adjust costs according to demand to sell all your stock.
As someone pointed out in the class, I am also curious about how Amazon deal with some holidays like Black Friday when most e-commerce websites including Amazon have a burst of visit? Does this mean Amazon has a much higher amount of over-provisioning servers than actually needed, which seems to be a waste of resource in most cases?
Fun fact: Japan once crashed Twitter by all tweeting about New Year's at the exact same time
http://www.tokyotimes.com/twitter-crash-during-japan-new-year-celebration/
@BigFish - Yes, I remember reading somewhere they were at like 25% capacity when not during Christmas. It made for really good DDOS protection. I think they rent that out via their cloud platforms (AWS) now though.
@BigFish: Yep! Kavyon mentioned that AWS came about out of trying to recoup some of their server costs during non-peak seasons. The cost of using AWS varies as Amazon's own demand on their servers varies (i.e. you pay a premium if you try to use AWS during the middle of the holiday shopping season).
@afzhang Thanks for answering. But my question is that if AWS renting out server during non-peak seasons, what should they do during peak seasons when there are a large demand of servers by users and Amazon itself? Does this imply that Amazon has to keep far more servers than needed during non-peak seasons even though they are renting out some of them.
@BigFish I don't know the exact answer to this, but I imagine that this is a case where market forces just make it work out. They should bump up the prices for computation power during peak periods to the point where services that DON'T see spikes in usage during those periods temporarily cut back to lower priority services. It's not ideal, and it's a bum deal for those who don't see the extra cost as worth the guaranteed computation power, but it should mean that the resources are being used at high efficiency year round. Recall that you rarely purchase physical processors from AWS; rather you purchase different degrees of performance and reliability.
More demand -> higher cost. Less demand -> lower cost. Adjust costs according to demand to sell all your stock.