lakesuperior_legacy/doc/notes/performance.txt

# Performance Benchmark Notes

## Environment

### Hardware

- Dell Precison M3800 Laptop
- 8x Intel(R) Core(TM) i7-4712HQ CPU @ 2.30GHz
- 12Gb RAM
- SSD

### Software

- Arch Linux OS
- glibc 2.26-11
- python 3.5.4
- lmdb 0.9.21-1
- db (BerkeleyDB) 5.3.28-3

### Sample Data Set

Modified Duchamp VIAF dataset (343 triples; changed all subjects to `<>`)

## Vanilla FCREPO 4.7

10K PUTs to new resources under the same container:

9'27" running time
0.057" per resource
3.4M triples total in repo at the end of the process

Retrieval of parent resource (~10000 triples), pipe to /dev/null: 3.64"

Database size: 3.3 Gb


## Sleepycat Back End Test

10K PUTs to new resources under the same container:

~18' running time
0.108" per resource
3.4M triples total in repo at the end of the process

Retrieval of parent resource, pipe to /dev/null: 3.6"

Database size: 1.2 Gb


## LMDB Back End Test

### Strategy #4

10K PUTs to new resources under the same container:

~29' running time
0.178" per resource
3.4M triples total in repo at the end of the process

Some gaps every ~40-50 requests, probably blocking transactions or disk
flush

Database size: 633 Mb

Retrieval of parent resource, pipe to /dev/null: 3.48"


### Strategy #5

10K PUTs to new resources under the same container:

29' running time
0.176" per resource
3.4M triples total in repo at the end of the process

Less gaps than strategy #4, however overall timing is almost identical. The
blocker seems to be somewhere else.

Database size: 422 Mb

Retrieval of parent resource, pipe to /dev/null: 7.5"


### After using triple methods rather than SPARQL for extract_imr

25' running time
0.155" per resource

Database size: 523 Mb

Retrieval of parent resource, pipe to /dev/null: 1.9"