Optimize the number of calls to fwrite() [#488288]

Comment #1

avpaderno

he/him

Italian

Brescia, 🇮🇹 🇪🇺

commented 11 June 2009 at 04:37

fwrite() is also called 6 times to output the sitemap data for a single link. Adding 6 times with the previously reported 5 times, plus one call more to output the closing sitemap tag, in a sitemap with just 2 links fwrite() is called 18 times; for a sitemap with 100 links, the function is called 606 times.
Using different code, in a sitemap with 100 links, the function could be called just 102 times. The difference with the actual code is 504 calls less, when the number of links is just 100.

Log in or register to post comments

Comment #2

avpaderno

he/him

Italian

Brescia, 🇮🇹 🇪🇺

commented 11 June 2009 at 04:43

In general, if n is the number of links in a sitemap chunk, the actual code calls fwrite (n + 1) * 6 times, while a different implementation would call the same function n + 2 times. For a sitemap chunk with 1000 links the number of times the function is called would be 6006 times versus 1002.

Log in or register to post comments

Comment #3

Anonymous (not verified) commented 11 June 2009 at 11:56

Status:	Active	» Postponed
Issue tags:		+remind

Changed you mind on the change you did? Check the 5.x code and you'll see that it sets a variable and then calls fwrite(). I'm leaning toward, not worth fixing at this point.

Log in or register to post comments

Comment #4

avpaderno

he/him

Italian

Brescia, 🇮🇹 🇪🇺

commented 11 June 2009 at 17:08

See the code I developed before you rolled it back to beta0; Drupal 5 branch changes were secondary, and it's not said I would promptly update it to the latest developed code for Drupal 6.

Log in or register to post comments

Comment #5

dave reid

he/him

English

Nebraska USA

commented 11 June 2009 at 18:13

I did some benchmarking on 6.x-2.x to see compare calling fwrite often vs one time per link. The test ran a full drupal bootstrap that just called xmlsitemap_regenerate(). My sitemap was split into two chunks and had only about 6500 links total. The sitemap index and chunks were generated twice for english and french languages. Ran ab -n 50 http://mysql.drupal6dev.local/test.php

OId code (fwrite called 6 times per link)

Requests per second:    0.36 [#/sec] (mean)
Time per request:       2741.419 [ms] (mean)
Time per request:       2741.419 [ms] (mean, across all concurrent requests)
Transfer rate:          0.21 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:  2605 2741  57.8   2752    2827
Waiting:     2597 2736  57.8   2746    2827
Total:       2605 2741  57.8   2752    2827

Optimized code (1 fwrite per link):

Requests per second:    0.39 [#/sec] (mean)
Time per request:       2546.178 [ms] (mean)
Time per request:       2546.178 [ms] (mean, across all concurrent requests)
Transfer rate:          0.23 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:  2370 2546  44.6   2544    2649
Waiting:     2365 2542  45.4   2542    2649
Total:       2370 2546  44.7   2544    2649

So with a fairly small sitemap, it resulted in about 200ms of an improvement which is a small, but worthwhile step that will become a greater difference as the sitemap grows larger. Thanks Kiam! I have implemented this change in 6.x-2.x with http://drupal.org/cvs?commit=224204.