Characterization of Existing Systems

1. Characterization of Existing Systems Puneet Chopra Ramakrishna Kotla Department of Computer Sciences, UT Austin

2. SEDA Applications • Component  SEDA stage • Stage  SEDA stage • Path  Application/user defined paths within applications • PE  Processor • Context  Event • Thread  Kernel threads in thread pool associated with each stage • Monitor  Stage queue length (implicit) • Policy Abstraction  • C  S : Mandatory • T  S : Thread pool + mechanism to adjust pool size based on queue back pressure. • PE T : Time slicing • C  T : FIFO/ user defined schedulers/ batching Department of Computer Sciences, UT Austin

3. StagedServer • Component  Stage • Stage  Stage • Path  Application built over SS library • PE  Processor • Context  Instance of Closure class • Thread  Associated with each PE • Monitor  Processor queue length (implicit) • Policy Abstraction  At different levels • C  S : Mandatory • T  S : Wavefront algorithm. • PE  S : T  S • PE T : Space Splicing • C  T : Cohort Scheduling Department of Computer Sciences, UT Austin

4. Cilk • Component  Non replicated node of dag • # of components = 2*(# of cilk fns) + # of syncs. • Stage  Spawned thread • Path  Runtime dag • PE  Processor • Context  Arguments passed to spawned threads. • Thread  Associated with each PE • Monitor  Work queue length (implicit) • Policy Abstraction  • C  S : Push bottom (fast clone) and Mandatory (slow clone) • T  S : Work Stealing. • PE  S : T  S • PE T : Space Splicing • C  T : LIFO Department of Computer Sciences, UT Austin

5. Hood • Component  Class inheriting HoodThread • Stage  Spawned thread • Path  Application • PE  Processor • Context  Arguments passed to HoodThreads. • Thread  HoodSerMach objects, specified at execution time. • Monitor  Work queue length (implicit) • Policy Abstraction  • C  S : Mandatory • T  S : Work Stealing. • PE  S : No explicit policy • PE T : Time Splicing • C  T : FIFO because only one context per stage Department of Computer Sciences, UT Austin

6. Coign • Component  COM object • Stage  Collection of COM objects instantiated on a machine. • Path  Application being partitioned • PE  Machine • Context  COM method invocation parameters • Thread  One or many associated with a PE. • Monitor  RTE imposed interface. • Policy Abstraction  • C  S : Mandatory • T  S : • PE  S : Space partitioning • PE T : • C  T : FIFO Department of Computer Sciences, UT Austin

7. JAWS • Component  Class methods • Stage  Framework, patterns, class libraries or objects. • Path  JAWS server. • PE  Processor • Context  Request context • Thread  Thread • Monitor  None to our knowledge. • Policy Abstraction  Service configurator pattern. • C  S : Mandatory • T  S : Multithreading • PE  S : • PE T : Time Splicing • C  T : FIFO + thread pool/thread per request/single thread Department of Computer Sciences, UT Austin

8. Click Modular Router • Component  Element • Stage  Element • Path  Packet flow path in router graph • PE  Processor (current design for uniprocessor) • Context  Packet • Thread  One, associated with the PE. • Monitor  Queue length, throughput, latency etc. • Policy Abstraction  Hierarchical CPU scheduling. • C  S : Mandatory • T  S : Task queue. • PE  S : No explicit policy • PE T : Space partitioning • C  T : any scheduler (FIFO etc) Department of Computer Sciences, UT Austin

9. Scout OS • Component  Router • Stage  Router • Path  Scout Path • PE  Processor • Context  Message • Thread  Thread • Monitor  None • Policy Abstraction  Thread scheduling on paths. • C  S : Mandatory • T  S : Multithreading. • PE  S : No explicit policy • PE T : Time/Space splicing • C  T : EDF and Fixed priority RR Department of Computer Sciences, UT Austin