How to achieve advertised benchmark performance? about nedb HOT 5 CLOSED

Hello,

The benchmarks should be run with indexing enabled to get the same resuts, so you should run node insert.js -i. That should do the trick !

Le 23 juin 2013 à 04:36, Jakub Szwacz [email protected] a écrit :

I've run nedb benchmark and was disappointed by the results. When in-memory-only all works perfect, but writing to file slows everything to disc I/O max speed.
Is it expected behaviour or some problem on my machine? I thought that the reason why copy of all database is kept in memory is to use it as a cache so disc I/O is not a problem. Something changed in the meantime?

Thanks for clarification.

Benchmark result:

node insert.js

Test with 10000 documents
Don't use an index
Don't use pipelining

Use a persistent datastore

INSERT BENCH - Begin profiling
INSERT BENCH - Begin inserting 10000 docs - 0ms (total: 0ms)
===== RESULT (insert) ===== 217 ops/s
INSERT BENCH - Finished inserting 10000 docs - 45.9s (total: 45.9s)
INSERT BENCH - Benchmark finished - 0ms (total: 45.9s)

node insert.js -m

Test with 10000 documents
Don't use an index
Don't use pipelining

Use an in-memory datastore

INSERT BENCH - Begin profiling
INSERT BENCH - Begin inserting 10000 docs - 0ms (total: 0ms)
===== RESULT (insert) ===== 11876 ops/s
INSERT BENCH - Finished inserting 10000 docs - 842ms (total: 842ms)
INSERT BENCH - Benchmark finished - 0ms (total: 842ms)
``'
—
Reply to this email directly or view it on GitHub.

from nedb.

Yes, I've tried it already. Little better but not good :)
Any ideas what could gone wrong are welcome.

node insert.js -i
----------------------------
Test with 10000 documents
Use an index
Don't use pipelining
Use a persistent datastore
----------------------------
INSERT BENCH - Begin profiling
INSERT BENCH - Begin inserting 10000 docs - 0ms (total: 0ms)
===== RESULT (insert) ===== 325 ops/s
INSERT BENCH - Finished inserting 10000 docs - 30.7s (total: 30.7s)
INSERT BENCH - Benchmark finished - 0ms (total: 30.7s)

from nedb.

(Louis here, on a friend's account) Yes I'm sorry that would have made sense for a benchmark on find, not insert. Seems like an issue with your filesystem. I tested on a few friends' computers as well as mine and we have much higher numbers, even though our laptops are not especially powerful. Do you have an exotic filesystem/OS combination ?

Additionally, I am working on a new persistence scheme that will be much faster, almost regardless of filesystem performance. t will take some time, I'll keep you posted!

from nedb.

Hah, I've found the culprit. It was my anti-virus which slowed down disc I/O by 10 fold (!?).
Sorry for the hassle. I had no clue modern hard drives are that fast, and thought 300 writes/sec is normal and expected result.

from nedb.

Wouldn't have thought about that one, interesting.

from nedb.