CannyLS is an embedded and persistent key-value storage optimized for random-access workload and huge-capacity HDD.
CannyLS mainly has following features:
- A local storage for storing objects that called as "lump":
- Basically, a lump is a simple key-value entry
- The distinctive properties are that the key is fixed length (128 bits) and suited for storing a relatively large size value (e.g., several MB)
- Provides simple functionalities:
- Basically, the only operations you need to know are
PUT,GETandDELETE - But it supports deadline based I/O scheduling as an advanced feature
- Basically, the only operations you need to know are
- Optimized for random-access workload on huge-capacity HDD (up to 512 TB):
- See Benchmark Results for more details about performance
- Aiming to provide predictable and stable read/write latency:
- There are (nearly) strict upper bounds about the number of disk accesses issued when executing operations
- One disk access when
PUTandDELETE, and two whenPUT
- One disk access when
- There are no background processings like compaction and stop-the-world GC which may block normal operations for a long time
- For eliminating overhead and uncertainty, CannyLS has no caching layer:
- It uses Direct I/O for bypassing OS layer caching (e.g., page cache)
- If you need any caching layer, it is your responsibility to implement it
- There are (nearly) strict upper bounds about the number of disk accesses issued when executing operations
- Detailed metrics are exposed using Prometheus
See Wiki for more details about CannyLS.