Machines Racks Cluster
DC Node Design
SMP: Symmetric Multi-processor
Set of CPU each with its own cache, sharing the main memory
Pros: high performance per node Cons: Expensive
Commodity Nodes
Pros: Equal performance as SMP, lower cost Cons: Fails more
More communication, better is SMP
Wimpy nodes
Lower end CPUs
Pros: lower cost, lower energy Cons: hard to use efficiently
Amdahl’s law
Higher number of threads higher communication cost, higher variance, harder to program, higher software cost, higher networking cost, lower utilization
Storage Design
Network Attached Storage (NAS)
Dedicated storage appliance
Pros: simple deployment, lower network overhead
Distributed storage: aggregate storage space from nodes in cluster
reduce cost by using cheap disks
Pros: lower cost, higher availability, higher performance, higher data locality Cons: higher network overhead, lower reliability