Active Record performance

[Thiago Lewin]

Hi,

  I am working on Goliath app version for Web Framework Benchmarks, just for fun.

  Some parts of benchmark ask for ORM version or if you decided not use it, you must have explicit it by adding a raw in front of it. I decided to keep both approaches in the same test, just for simplicity, but when I include the ORM version (ActiveRecord) I realized the general performance was compromised.

  I create a simple example to investigate the process: https://gist.github.com/tlewin/6587233

  When I run a simple benchmark (ab -n 1000 -c 10 http://localhost:9000/), I get the following results:

Concurrency Level:      10

Time taken for tests:   3.214 seconds

Complete requests:      1000

Failed requests:        0

Write errors:           0

Total transferred:      104208 bytes

HTML transferred:       11022 bytes

Requests per second:    311.18 [#/sec] (mean)

Time per request:       32.136 [ms] (mean)

Time per request:       3.214 [ms] (mean, across all concurrent requests)

Transfer rate:          31.67 [Kbytes/sec] received

  

  When I uncomment the second line, I mean, when I just include the ActiveRecord gem, I get the following results:

Concurrency Level:      10

Time taken for tests:   6.300 seconds

Complete requests:      1000

Failed requests:        0

Write errors:           0

Total transferred:      104312 bytes

HTML transferred:       11033 bytes

Requests per second:    158.74 [#/sec] (mean)

Time per request:       62.996 [ms] (mean)

Time per request:       6.300 [ms] (mean, across all concurrent requests)

Transfer rate:          16.17 [Kbytes/sec] received 

  I couldn't understand what is going on. I start to read the AR code, but it is not an easy task and will take some time.

  Does anyone have a clue what is happening?

  I ran the benchmarks with Ruby 2.0.0/ Goliath 1.0.3 on a macbook pro.

Regards,

Thiago

[Ilya Grigorik]

That's bizarre. Profile it, see where the times goes.

Also, make sure you're running goliath server with "prod" flag (-e prod) --
otherwise you'll be reloading code on every request.

ig

--
You received this message because you are subscribed to the Google Groups "Goliath.io" group.
To unsubscribe from this group and stop receiving emails from it, send an email to goliath-io+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Thiago Lewin]

I was in dev mode :-( 

I repeated the benchmark several times, trying to remove any noise, now using "ab -n 10000 -c 100 http://localhost:9000" and the version without AR inclusion is about 15% faster.

Whitout AR:

Concurrency Level:      100

Time taken for tests:   5.092 seconds

Complete requests:      10000

Failed requests:        0

Write errors:           0

Total transferred:      1040000 bytes

HTML transferred:       110000 bytes

Requests per second:    1963.96 [#/sec] (mean)

Time per request:       50.918 [ms] (mean)

Time per request:       0.509 [ms] (mean, across all concurrent requests)

Transfer rate:          199.46 [Kbytes/sec] received

With AR:

Concurrency Level:      100

Time taken for tests:   6.035 seconds

Complete requests:      10000

Failed requests:        0

Write errors:           0

Total transferred:      1040624 bytes

HTML transferred:       110066 bytes

Requests per second:    1656.87 [#/sec] (mean)

Time per request:       60.355 [ms] (mean)

Time per request:       0.604 [ms] (mean, across all concurrent requests)

Transfer rate:          168.38 [Kbytes/sec] received

I will profile it in detail, as you suggested, but my guess is the overhead of number of objects created by AR implies in more GC calls.

Many thanks!

Thiago