[Plugin: DB Cache] Queries are reduced but load time is not

it uses serialize/unserialize and fwrite/fread to operate with cache.

i wrote this plugin special for one of my sites with much widgets that needed always actual information to be displayed. so with this plugin i’ve got 160 -> 2 queries and 4 second -> 0.5 seconds.

at the moment i’m checking work of plugin with different plugins. possibly increasing of genereation time is because of not supporting one of plugins. so solving problems. hope that it will be fine on every instalation of wordpress soon

True, but what I was saying is it would be helpful to measure the cpu load of serializing and unserializing the query results. If it takes more cpu to perform that i/o than it does to query and parse the database results, then under load you still really wouldn’t have a benefit and would still need something like super cache to survive a digg hit.

I like the plugin, we just need someone to do some load testing to see how it performs under full load. I’m going to try if I get some time, but I’m certainly not an expert at it.

Ok, I’ll watch for the new version and try to do some really basic testing using ab.

Ok, I did some basic benchmarking. Keep in mind this is on a vps that doesn’t have much power, but it suitable for a few low volume sites. I’d be curious to see where the numbers went on a machine with a lot more cpu and more memory. If I have time, I’ll benchmark the site on my laptop and see what the numbers there are.

My method:
I ran the benchmark’s twice. I first ran the site with no caching. Then I ran the site with db-cache involved. Then, just to get a comparison with one of the other caching techniques, I ran it with hyper cache. Apache was restarted between each run to clean out it’s memory and worker processes. I also “primed the cache” between each run by navigating to a few pages.

The summary is that db-cache 0.6 is better than no cache, but still trails hyper cache by a wide margin. Again, I’d be curious to see if the gap narrows on a machine with a lot of cpu. Db cache is being slowed down because it still has to execute the php scripts and serialize and deseriaze the data whereas hyper cache is basically serving pages without a database hit. I’m not being negative about db cache, just trying to account for the difference. It might be good to see if there is a way to optimize the saving and loading of queries. That might be a place for optimization. Having all the php scripts execute each time though is going to make it hard to match caching techniques that serve a written page like hyper or super cache (which obviously have their own downsides as well). I’m not being negative, I just think you have to be honest that db cache is never going to match the speed of something like super cache because of the php code that’s still running. Still, db cache could be a good alternative for people that don’t want the complexity of super cache.

For the curious, here’s the details:
******* No cache – first run:
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking http://www.samuelclough.com (be patient)
Finished 108 requests

Server Software: Apache/2.0.55
Server Hostname: http://www.samuelclough.com
Server Port: 80

Document Path: /
Document Length: 109441 bytes

Concurrency Level: 10
Time taken for tests: 30.468 seconds
Complete requests: 108
Failed requests: 0
Write errors: 0
Total transferred: 12433486 bytes
HTML transferred: 12402598 bytes
Requests per second: 3.54 [#/sec] (mean)
Time per request: 2821.149 [ms] (mean)
Time per request: 282.115 [ms] (mean, across all concurrent requests)
Transfer rate: 398.51 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 34 59 44.1 45 321
Processing: 797 2535 1294.5 2377 7395
Waiting: 138 557 378.2 438 2338
Total: 838 2594 1297.6 2416 7435

Percentage of the requests served within a certain time (ms)
50% 2416
66% 2888
75% 3102
80% 3432
90% 4356
95% 5378
98% 6016
99% 6186
100% 7435 (longest request)

******** No cache – second run:
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking http://www.samuelclough.com (be patient)
Finished 129 requests

Server Software: Apache/2.0.55
Server Hostname: http://www.samuelclough.com
Server Port: 80

Document Path: /
Document Length: 109441 bytes

Concurrency Level: 10
Time taken for tests: 30.168 seconds
Complete requests: 129
Failed requests: 0
Write errors: 0
Total transferred: 14707789 bytes
HTML transferred: 14671357 bytes
Requests per second: 4.28 [#/sec] (mean)
Time per request: 2338.594 [ms] (mean)
Time per request: 233.859 [ms] (mean, across all concurrent requests)
Transfer rate: 476.11 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 35 47 13.8 44 181
Processing: 663 2182 1641.6 1758 15478
Waiting: 136 625 636.5 446 5012
Total: 707 2229 1643.3 1804 15522

Percentage of the requests served within a certain time (ms)
50% 1772
66% 2157
75% 2564
80% 2902
90% 4266
95% 4846
98% 5059
99% 6815
100% 15522 (longest request)

********** db cache – first run:
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking http://www.samuelclough.com (be patient)
Finished 159 requests

Server Software: Apache/2.0.55
Server Hostname: http://www.samuelclough.com
Server Port: 80

Document Path: /
Document Length: 109577 bytes

Concurrency Level: 10
Time taken for tests: 30.286 seconds
Complete requests: 159
Failed requests: 0
Write errors: 0
Total transferred: 17969959 bytes
HTML transferred: 17925607 bytes
Requests per second: 5.25 [#/sec] (mean)
Time per request: 1904.792 [ms] (mean)
Time per request: 190.479 [ms] (mean, across all concurrent requests)
Transfer rate: 579.43 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 35 53 35.0 44 318
Processing: 685 1762 1069.8 1415 6025
Waiting: 130 412 322.1 270 1827
Total: 725 1815 1069.3 1462 6069

Percentage of the requests served within a certain time (ms)
50% 1446
66% 1965
75% 2210
80% 2393
90% 3463
95% 3886
98% 5333
99% 5560
100% 6069 (longest request)

********** db cache – second run:
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking http://www.samuelclough.com (be patient)
Finished 174 requests

Server Software: Apache/2.0.55
Server Hostname: http://www.samuelclough.com
Server Port: 80

Document Path: /
Document Length: 109577 bytes

Concurrency Level: 10
Time taken for tests: 30.068 seconds
Complete requests: 174
Failed requests: 0
Write errors: 0
Total transferred: 19443356 bytes
HTML transferred: 19395836 bytes
Requests per second: 5.79 [#/sec] (mean)
Time per request: 1728.065 [ms] (mean)
Time per request: 172.807 [ms] (mean, across all concurrent requests)
Transfer rate: 631.48 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 35 49 70.3 42 967
Processing: 667 1636 729.2 1508 4247
Waiting: 127 401 234.3 352 1248
Total: 746 1686 727.6 1570 4286

Percentage of the requests served within a certain time (ms)
50% 1570
66% 1806
75% 2023
80% 2222
90% 2710
95% 3291
98% 3730
99% 3919
100% 4286 (longest request)

************ hyper cache – first run:
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking http://www.samuelclough.com (be patient)
Finished 302 requests

Server Software: Apache/2.0.55
Server Hostname: http://www.samuelclough.com
Server Port: 80

Document Path: /
Document Length: 107579 bytes

Concurrency Level: 10
Time taken for tests: 30.038 seconds
Complete requests: 302
Failed requests: 0
Write errors: 0
Total transferred: 32768332 bytes
HTML transferred: 32702922 bytes
Requests per second: 10.05 [#/sec] (mean)
Time per request: 994.629 [ms] (mean)
Time per request: 99.463 [ms] (mean, across all concurrent requests)
Transfer rate: 1065.34 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 38 126 113.9 105 1133
Processing: 389 844 225.6 809 2246
Waiting: 39 121 118.4 98 1197
Total: 430 970 239.6 930 2289

Percentage of the requests served within a certain time (ms)
50% 930
66% 974
75% 998
80% 1024
90% 1195
95% 1381
98% 1804
99% 1941
100% 2289 (longest request)

************** hyper cache – second run:
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking http://www.samuelclough.com (be patient)
Finished 298 requests

Server Software: Apache/2.0.55
Server Hostname: http://www.samuelclough.com
Server Port: 80

Document Path: /
Document Length: 107579 bytes

Concurrency Level: 10
Time taken for tests: 30.075 seconds
Complete requests: 298
Failed requests: 0
Write errors: 0
Total transferred: 32393644 bytes
HTML transferred: 32329289 bytes
Requests per second: 9.91 [#/sec] (mean)
Time per request: 1009.228 [ms] (mean)
Time per request: 100.923 [ms] (mean, across all concurrent requests)
Transfer rate: 1051.85 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 40 120 135.0 91 1164
Processing: 508 862 214.8 806 1867
Waiting: 40 122 110.1 95 936
Total: 606 982 269.7 889 2362

Percentage of the requests served within a certain time (ms)
50% 889
66% 958
75% 1074
80% 1188
90% 1372
95% 1503
98% 1857
99% 1942
100% 2362 (longest request)

Ok guys, I have a question. I undestand that hypercache and dbcache are tow different ways of fixinf the same issue.

My question, can/should they be use together? or it will be like double effort to use them both?

thanks

it may be possible that the cpu overhead of parsing the cache is equal to the i/o overhead of calling the database if you are dealing with small queries. I’d be interested to get some input from others on this.

This is seen very easily with a define('SAVEQUERIES', true); (as per http://codex.wordpress.org/Editing_wp-config.php#Save_queries_for_analysis ) and a simple hack on db-module.php (and also to the copy in /wp-content/db.php). Add the following somewhere near line 720:

if ( defined('SAVEQUERIES') && SAVEQUERIES )
	$this->queries[] = array( $query, $this->timer_stop(), $this->get_caller(), "Cache hit" );

(Posterd, you may like to commit this hack to your code.)

Then load a page in two browser windows, one minute apart.

It will be evident that DB Cache shaves milliseconds on some queries such as queries with JOINs.