The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines by Luiz André Barroso and Urs Hölzle of Google introduces the notion of WSC’s (warehouse-scale computers); an Internet service that spans thousands of computing nodes and the associated hardware that goes along with it as a single computing unit.
The book goes into the architecture of such WSC’s, workloads & software infrastructure, hardware building blocks, datacenter basics, energy & power efficiency, costs, dealing with failures & repairs and the challenges in designing such machines.
Some interesting points from the book:
- programmers must be aware of the relatively scarce cluster-level bandwidth resources and try to exploit rack-level networking locality
- distribution of peak power usage – CPU’s 33%, DRAM 30%
- cost-efficiency metrics for a large SMP server and a low-end, PC-class server
- At the moment, the sweet spot for many large-scale services seems to be at the low-end range of server-class machines
- Energy management being a key issue. Focus on reducing all energy-related costs capital, operating expenses, and environmental impacts.
- The average real-world datacenter and the average server are far too inefficient, mostly because efficiency has historically been neglected and has taken a backseat relative to reliability, performance, and capital expenditures. As a result, the average WSC wastes two thirds or more of its energy
- As a rule of thumb, most large datacenters probably cost around $12–15/W to build and smaller ones cost more
- The total cost of a server will be primarily a function of the power it consumes, and the server’s purchase price will matter less. In other words, over the long term, the datacenter facility costs (which are proportional to power consumption) will become a larger and larger fraction of total cost
You can download the book here.
