TECH

1
Scaling out Google Style
Sydney BarCamp 07 Joel Pobar joelpobar_at_gmail.com
http//callvirt.net/blog/
2
Agenda

• Parallelism Concurrency
• Hardware
• Programming Models
• MapReduce
• Experiments to date

3
Definitions

• Concurrency
• Dijkstra -- Concurrency occurs when two or more
  execution flows are able to run simultaneously
• Parallelism
• Simultaneous execution of the same task!

4
HardwarePerformance The Multi-Core era

• Processors dont get way faster anymore You
  just get a whole lot more slow ones!!!

5
Hardware

• 90?65?45 nm lithography advances
• Twice, twice again as many (faster) transistors
• Slower wires maxed out thermal envelope?
  slower CPU frequency scaling
• Same freq more cores more cache RAM? same
  cache/core and compute/core
• Architecture advances
• (Hardware) multithreading
• Optimizing for throughput, power
• System-on-a-Chip integration interconnects,
  shared caches, I/O and DRAM controllers
• Intel talks 80 cores!?

6
Programming ModelsClient side

• Shared memory, threads and locks
• Most used, most disastrous
• Synchronisation is costly shared memory
  accesses across multiple CPUs doesnt scale
  (cache misses etc)
• Tough heisenbugs
• Loop parallelism OpenMP
• Message Passing CCR, MPI, Erlang
• Functional Languages Implicit, no shared state
• Software Transactional Memory
• IMO Most likely to solve the problem

7
Programming ModelsServer side

• Server per-client work-unit parallelism
• Web server implicit request parallelism
• SQL Implicit data parallelism
• Scale out possible, but bottlenecks can occur at
  layer boundaries
• Typically hard to scale out to lots of machines
• Clusters (Beowulf, Windows HPC)
• Grid Computing/Cycle Stealing (Sun Grid, G2,
  Alchemi)
• Map/Reduce (Hadoop Java, Open Source, Google
  MapReduce Not available

8
Agenda

• Parallelism Concurrency
• Hardware
• Programming Models
• MapReduce
• Experiments to date

9
MapReduce

• Nice functional programming model (similar to
  Googles MapReduce model)
• Scheduling, latency, file system, resource
  management
• Things to think about Hyperthreading,
  Programming model, code distribution, security,
  resource management for dummies, automatic
  scheduling and tuning
• What we did

10
MapReduce.NET experiement
11
Contact

• Joel Pobar
• Brisbane based
• 0410 443 469
• joelpobar_at_gmail.com