Lesson 15.2: Multiprocessing – Thread-Level Parallelism to Warehouse-Scale

Scalable Architecture

is an architecture that can scale up to meet increased work loads

More Processor

scalability to _ _ is the key issue

Cache Coherence

processors may see different values through their caches

Cache Coherence

in a coherent memory system read of a data item returns the most recently written value

Cache Coherence

2 important mechanism (1)

Memory Consistency

2 important mechanism (2)

Cache Coherence

defines behavior of reads and writes to the same memory location

Recently Written Value

cache coherence, all reads by any processor must return most _ _ _

Same Location, 2 Processor

cache coherence, writes to the _ _ by any _ _ are seen in same order by all processors

Memory Consistency

defines behavior of reads and writes with respect to access to other memory locations

Different Locations

memory consistency, serialized writes _ _

Coherence

one location over time

Consistency

ordering across locations

Migration

movement of data

Replication

multiple copies of data

Snooping

each core tracks sharing status (thru a shared broadcast bus) of each of its blocks

Distributed, Low Scalability

snooping, limited to a small multi-core processor

Directory Based

sharing status of each block kept in a single location (thru a centralized controller)

Centralized, High Scalability

directory based, used in massive multi-socket servers / supercomputers

Warehouse Scale Computers

meaning of WSCs

Single Macro-computer

WSCs Scale and Scope, massive data centers functioning as a _ _

RLP

WSCs exploit natural request-level parallelism, handling millions of independent user queries concurrently

Request Level Parallelism

meaning of RLP

Dependability via Redundancy

hardware failures are treated as a statistical certainty

Dependability via Redundancy

reliability is engineered through software replication and redundant commodity servers rather than hyper-expensive fault-tolerant mainframes

PUE

WSCs energy efficiency, measured via _ balancing massive multi-stage power distribution and industrial cooling

Power Usage Effectiveness

meaning of PUE

Total Facility Power / IT Equipment Power

formula of PUE

Communication Latency

WSCs Network Hierarchy, multi-tiered switching topologies to minimize _ _ across tens of thousands of nodes