Skip to content

Latest commit

 

History

History
44 lines (37 loc) · 6.5 KB

README.md

File metadata and controls

44 lines (37 loc) · 6.5 KB

This is a public mirror of a private repository from which certain files have been removed. The repo is frequently forced pushed to.

Language comparison

Savina Benchmark

List of Savina benchmarks.

# Name Symbol Feature or Pattern being measured Source
micro-benchmarks
1 Ping Pong PP Message delivery overhead Scala
2 Counting Actor COUNT Message passing overhead Theron
3 Fork Join (throughput) FJT Messaging throughput JGF , ourselves
4 Fork Join (actor creation) FJC Actor creation and destruction JGF , ourselves
5 Thread Ring THR Message sending; Context switching between actors Theron
6 Chameneos CHAM Contention on mailbox; Many-to-one message passing Haller
7 Big BIG Contention on mailbox; Many-to-Many message passing BenchErl
concurrency
8 Concurrent Dictionary CDICT Reader-Writer concurrency; Constant-time data structure Ourselves
9 Concurrent Sorted Linked-List CSLL Reader-Writer concurrency; Linear-time data structure Shirako et al.
10 Producer-Consumer with Bounded Buffer PCBB Multiple message patterns based on Join calculus Sulzmann et al.
11 Dining Philosophers PHIL Inter-process communication; Resource allocation Wikipedia
12 Sleeping Barber SBAR Inter-process communication; State synchronization Wikipedia
13 Cigarette Smokers CIG Inter-process communication; Deadlock prevention Wikipedia
14 Logistic Map Series LOGM Synchronous Request-Response with non-interfering transactions Wikipedia
15 Bank Transaction BTX Synchronous Request-Response with interfering transactions Ourselves
parallrlism
16 Radix Sort RSORT Static Pipeline; Message batching StreamIT
17 Filter Bank FBANK Static Pipeline; Split-Join Pattern StreamIT
18 Sieve of Eratosthenes SIEVE Dynamic Pipeline GPars
19 Unbalanced Cobwebbed Tree UCT Non-uniform load; Tree exploration Zhao and Jamali
20 Online Facility Location OFL Dynamic Tree generation and navigation Ourselves
21 Trapezoidal Approximation TRAPR Master-Worker; Static load-balancing Stage
22 Precise Pi Computation PIPREC Master-Worker; Dynamic load-balancing Ourselves
23 Recursive Matrix Multiplication RMM Uniform load; Divide-and-conquer style parallelism Ourselves
24 Quicksort QSORT Non-uniform load; Divide-and-conquer style parallelism Ourselves
25 All-Pairs Shortest Path APSP Phased computation; Graph exploration Ourselves
26 Successive Over-Relaxation SOR 4-point stencil computation SOTER
27 A-Star Search ASTAR Message priority; Graph exploration Ourselves
28 NQueens first N solutions NQN Message priority; Divide-and-conquer style parallelism Ourselves