Semi-dedicated benchmarks

[lobos]

There is a free php benchmark script available here:

http://www.free-webhosts.com/php-benchmark-script.php

I ran it on my old dreamhost account and go this:

Test #1 completed in 72 ms.
Test #2 completed in 39 ms.
Test #3 completed in 105 ms.
Test #4 completed in 56 ms.
Test #5 completed in 44 ms.
Test #6 completed in 48 ms.
Test #7 completed in 41 ms.
Test #8 completed in 39 ms.
Test #9 completed in 46 ms.
Test #10 completed in 80 ms.
Test #11 completed in 68 ms.
Test #12 completed in 42 ms.

Lowest time: 39 ms , Highest time : 105 ms
Average of all 12 times: 57 ms
Average of middle 10 times: 54 ms

Then I ran it on my jag semi dedicated account and got this:

Test #1 completed in 75 ms.
Test #2 completed in 72 ms.
Test #3 completed in 72 ms.
Test #4 completed in 75 ms.
Test #5 completed in 70 ms.
Test #6 completed in 72 ms.
Test #7 completed in 80 ms.
Test #8 completed in 70 ms.
Test #9 completed in 70 ms.
Test #10 completed in 71 ms.
Test #11 completed in 70 ms.
Test #12 completed in 72 ms.

Lowest time: 70 ms , Highest time : 80 ms
Average of all 12 times: 72 ms
Average of middle 10 times: 72 ms

Here is a list of benchmarks from FREE php hosts:

http://www.free-webhosts.com/php-hosting-comparison.php

What I am wondering is why my semi dedicated account is not kicking all there asses...

I would like to see the results from other semi-dedicated account holders as well...

[lobos]

Well does anyone have any advice as to why free hosting sites and Dreamhost (yuck!) display consistantly better benchmark result than my semi dedicated account?

Thanks.

[Gwaihir]

Consistantly? I see you post just one moment in time. Was the server working hard for other users at that time, or was it pretty much idle? You post not even the most basic load numbers for reference, so I guess you didn't check.

Suppose it was completely idle (though I can't imagine it was). What good is it to you to measure how fast it can do some single thread CPU oriented tasks? Why do you presume your results relate in any usefull way to how it performs running lots of concurrent requests involving mulitple threads, memory performance, disk I/O, etc.?

I am not in favour of folks running benchmarking suites in a shared production environment. Your desire to create some meaningless numbers is eating into real world, production performance for everyone else on the server.

[JPC-Greg]

Hard to tell really but if you research some more and let me know the details we can check out whats being tested and compare. Try to find out what processors, ram, disks, are being used in those other machines.

Maybe its also worth noting the tight group of results on our server vs the erratic results elsewhere. Most free webhosts have nothing to offer, no custom environments, no special libraries or packages loaded, just bare bones builds of php, apache, etc. Perhaps its the loading of the very customized env we run here thats eating some ram and resulting in lower numbers on this script? I dont know, and maybe thats way out in left field. We need to know what they are running first.

Along with the hardware specs , and most importantly, get some phpinfo details from each. Pm or email me the data and ill check it all out.

[wuurp]

Test #1 completed in 90 ms.
Test #2 completed in 91 ms.
Test #3 completed in 86 ms.
Test #4 completed in 90 ms.
Test #5 completed in 88 ms.
Test #6 completed in 88 ms.
Test #7 completed in 88 ms.
Test #8 completed in 87 ms.
Test #9 completed in 89 ms.
Test #10 completed in 89 ms.
Test #11 completed in 89 ms.
Test #12 completed in 88 ms.

Lowest time: 86 ms , Highest time : 91 ms
Average of all 12 times: 89 ms
Average of middle 10 times: 89 ms


That's on an Enterprise level VPS. Not sure what it's actually testing as load through out the test was practically 0. On a dedicated (low end) the values were all right about 101.

Hmm.....

[Gear]

JaguarPC semi-dedicated (Argus) 6:45AM EST

Test #1 completed in 59 ms.
Test #2 completed in 60 ms.
Test #3 completed in 59 ms.
Test #4 completed in 59 ms.
Test #5 completed in 59 ms.
Test #6 completed in 59 ms.
Test #7 completed in 58 ms.
Test #8 completed in 58 ms.
Test #9 completed in 59 ms.
Test #10 completed in 59 ms.
Test #11 completed in 262 ms.
Test #12 completed in 64 ms.

Lowest time: 58 ms , Highest time : 262 ms
Average of all 12 times: 76 ms
Average of middle 10 times: 60 ms

[Vin DSL]

Heh!

Always a fart in a carload...

Test #1 completed in 44 ms.
Test #2 completed in 44 ms.
Test #3 completed in 44 ms.
Test #4 completed in 44 ms.
Test #5 completed in 43 ms.
Test #6 completed in 56 ms.
Test #7 completed in 138 ms.
Test #8 completed in 44 ms.
Test #9 completed in 46 ms.
Test #10 completed in 46 ms.
Test #11 completed in 45 ms.
Test #12 completed in 47 ms.

Lowest time: 43 ms , Highest time : 138 ms
Average of all 12 times: 53 ms
Average of middle 10 times: 46 ms

Um...

I used PHP 5. Is that cheating? They got that at free sites, right?

[Ron]

Hmm, let me rewrite that in C and compile it...

[wuurp]

Hmm, let me rewrite that in C and compile it...

But isn't the idea of this benchmark to see how PHP performs?

[the_ancient]

But isn't the idea of this benchmark to see how PHP performs?

dont worry about Ron, he hates PHP, People that write PHP are low Class, wanna bees

[Ron]

dont worry about Ron, he hates PHP, People that write PHP are low Class, wanna bees

That's true, but it's not the point. Seriously, I've enjoyed playing with PHP, but it is just a flash in the pan. The next great thing is coming.

I am seriously looking into some PHP compilers and/or pre-compilers to use if I move to a VPS or dedicated solution, and will begin to write all of my backends in C immediately if I can get the stupid MySQL libraries to compile in. C is *literally* 100 times faster from the command line. I hope the same is true when called from Apache.

If Masood reads this; have you looked into making (the now FREE) roadsend available in our hosting environment to reduce the loads on the machines and "cram" more clients per box?

[the_ancient]

That's true, but it's not the point. Seriously, I've enjoyed playing with PHP, but it is just a flash in the pan. The next great thing is coming.

I am seriously looking into some PHP compilers and/or pre-compilers to use if I move to a VPS or dedicated solution, and will begin to write all of my backends in C immediately if I can get the stupid MySQL libraries to compile in. C is *literally* 100 times faster from the command line. I hope the same is true when called from Apache.

If Masood reads this; have you looked into making (the now FREE) roadsend available in our hosting environment to reduce the loads on the machines and "cram" more clients per box?

Ya, php is just a "flash in the pan"

[Vin DSL]

Heh!

I was looking at that benchmark code, and added a couple of lines to optimize the iterator...

Test #1 completed in 45 ms.
Test #2 completed in 47 ms.
Test #3 completed in 44 ms.
Test #4 completed in 44 ms.
Test #5 completed in 45 ms.
Test #6 completed in 44 ms.
Test #7 completed in 44 ms.
Test #8 completed in 44 ms.
Test #9 completed in 45 ms.
Test #10 completed in 45 ms.
Test #11 completed in 49 ms.
Test #12 completed in 44 ms.

Lowest time: 44 ms , Highest time : 49 ms
Average of all 12 times: 45 ms
Average of middle 10 times: 45 ms

I think this makes the result[s] more accurate, yes?

*edit*

BTW, I'm running on SDX (Semi-Dedicated)...

[Gwaihir]

But isn't the idea of this benchmark to see how PHP performs?

What is PHP to you? To me it isn't a calculator, which this test makes out of it, but a tool to process data, giving each user an up to date, personalised answer to his requests. That deals with some calculations, sure, but far more with opening and closing files, working with a database, etc., and doing all that for a number of users at the same time.

If a freehost builds a simple server, with a P-IV 3.0Ghz processor (or perhaps even a Celeron), one (sata) disk and a tiny bit of memory, it can perform such a lame test just as well as JagPC's SDX-machines, perhaps even slightly better (as dual core dual opterons are optimized for over all performance, not single thread performance). However, if you give it real world tasks, its lack of I/O speed, memory, and ability to run mulitple threads at once will cause results to plummet, compared to these more serious servers. Not to mention how much easier such a budget system can loose your data (takes but one disk crash), but well.. "free" doesn't come with guarantees, right?

Really, if you want serious benchmarking, go find serious benchmarks first. If you or lobos want the illusion of a super fast server, or most value for least money, then why do you have an Enterprise VPS or an SDX account, rather than something at a free host? Can't ever beat "free" then, can you?


Ron: go right ahead. It should be an interesting comparison between the compilers. PHP code is compiled too: when the script is run, it is compiled first and then executed. These tests are NOT timing the compile step, that's done when their clock starts ticking. One would assume that a good C compiler is more efficient than PHP's compiler. PHP's compiler is far more pressed for time, as it has to do his compiling with the end user waiting on it, so one would assume some short cuts in compiling in order to obtain best over all (compile + execute) effiency.

However, if the difference isn't as significant as you'd like, you can save yourself the trouble of doing everything in C and just use precompiled PHP. That's as simple as installing and tweaking the proper op-code cache on that VPS / dedicated of yours.

[Ron]

Huh? Are you saying that PHP is compiled in real time into native machine code and then executed? I wouldn't think so, but hey, IDKE.