This reading talks a lot about the costs of maintaining a datacenter.
Chapter 3 starts off by rationalizing the use of cheap servers over the more expensive servers in datacenters. They show that the increase in performance by using expensive servers don't justify the cost of these servers, and it is much more cost-efficient to buy more low-end servers. Of course, with servers that are too cheap cloud users will notice increases in response time, which is not desirable. Basically, it is a tradeoff.
There is one small thing that I wanted to mention about this chapter, is that the 100 microsecond estimate for network latencies seems really small. In a warehouse full of computers, it seems going from one end to the other end would take a lot longer. Also, the services would probably run on VMs and that would add to the increase in network latency, so I think the 100 microsecond estimate may be an underestimate.
Chapter 4 talks about the utilities required to run a datacenter. It talks about the various types of redundancies that are available for datacenters. The reading talks of how power is distributed to the datacenter using redundant PSUs. The reading shows how typical datacenters remove hot air from datacenters.
Chapter 7 talks about the availability of datacenters. Once you have thousands of very low-end machines, it is very likely that these machines will break. So, multiple topics surrounding this problem are discussed. Something, I found interesting was that it seems that number of restarts required among machines in the datacenter is quite high. Even the ones that are supposed to run well still require around 2-3 restarts a year. I was not aware that these machines had to be restarted so often. It's really interesting that even though some machines have to be restarted, services continue uninterrupted.
Thanks for the post. I liked it. Keep going I follow you.
ReplyDeleteVee Eee Technologies| Vee Eee Technologies|