conc-map-bench uses the bustle benchmarking harness. This is a port of the well regarded libcuckoo benchmark.

Implementations benchmarked

• RwLock Std HashMap
• CHashMap
• Contrie
• Flurry
• DashMap
• Evmap
• Crossbeam Skiplist
• LeapMap

> cargo tree | rg 'dashmap|chashmap|contrie|flurry|evmap|crossbeam-skiplist'

├── chashmap v2.2.2
├── contrie v0.1.4
├── crossbeam-skiplist v0.0.0 (https://github.com/crossbeam-rs/crossbeam#e1d64600)
├── dashmap v5.0.0
├── evmap v10.0.2
├── flurry v0.3.1

Workloads

Ready Heavy

A read heavy model with few inserts, removals and updates. Models caching of data in places such as webservers and disk page caches.

read   98%
insert  1%
remove  1%

Exchange

Insert and remove heavy model that replicates a scenario where the map is used to exchange data.

read    10%
insert  40%
remove  40%
update  10%

Rapid Grow

An insert heavy model that replicates load in a scenario where the map is used to gather large amounts of data under a short burst.

read    5%
insert 80%
remove  5%
update 10%

How to run it?

mv results results.bk
./scripts/bench.bash
./scripts/plot.bash

Results

OS: Ubuntu 20.4 CPU: AMD Threadripper 3950x

The CPU has 16 cores and 32 threads.

See the results/ directory.

Ready Heavy (fx hasher)

Exchange (fx hasher)

NOTE: the flurry case fails because of the OOM killer.

Rapid Grow (fx hasher)

Read Heavy (std hasher)

Exchange (std hasher)

NOTE: the flurry case fails because of the OOM killer.

Rapid Grow (std hasher)