php: sending 10% of your traffic

Question

My code:

$random_number = mt_rand(1,100);

if ($random_number <= 10){
    header('Location: http://www.site1.com');
} else {
    header('Location: http://www.site2.com');
}

exit;

Needless to say, it seems to work if you have a lot of traffic, but when you are down to around 1000 uniques a day, the percentage hovers around 35 to 40%.

according to your code, users should only be redirected (on an average of) 9% of the time due to the `<` which really should be a `<=`. although, tbh, why aren't you just selecting a random number 1 through 10 and checking it against the number 1? — jeremy, Aug 14 '12 at 21:29
Why don't you just store a counter on the webserver? The counter could be stored in a file, database, memcached, redis, or whatever floats your boat. Just count to 10, redirect, and reset the counter afterwards. — Rem.co, Aug 14 '12 at 21:30
@RemcoOverdijk, you believe the percentage would be better if I made it 10% over a range of 1, 10? — NiteRain, Aug 15 '12 at 00:15
@NiteRain, No, I'm trying to tell you that random numbers are, well.. random. So there's no reliable way of getting 10% out of any range of RANDOM numbers. Random numbers are not evenly (or even merely so) distributed in any way so for your purposes it is explicitly wrong to use them or rely on them. What I was trying to say is that you have to count the actual visitors and store the count on the server. If count < 10, redirect to site B; If count == 10, redirect to site A, reset counter to 0 and store it. — Rem.co, Aug 15 '12 at 00:25

score 1 · Accepted Answer · answered Aug 14 '12 at 21:30

1

For a more accurate distribution of traffic use a load balancer or increment a counter in your code.

Here is a nice article on why mt_rand isn't so random

answered Aug 14 '12 at 21:30

Martin

6,632
4
25
28

That article says that they're it's random enough for cryptographic use. Cryptographers aren't worried about the distribution of the data, but how easy it is to predict the sequence. It should be good enough for monte carlo use like this. – Barmar Aug 14 '12 at 21:40
The link is broken – Abraham Jan 27 '23 at 14:29

score 0 · Answer 2 · answered Aug 14 '12 at 21:42

0

If you're looking to do some A/B testing, there's a better way than blindly choosing at random and it's not really more complex.

answered Aug 14 '12 at 21:42

Sean McSomething

6,376
2
23
28

1

@aug - Looks like it's gone for good. It was a link to an article about the "Multi-armed Bandit" - https://support.google.com/analytics/answer/2844870?hl=en is the best replacement I can find in 5 minutes of searching. – Sean McSomething Jan 29 '16 at 09:22

php: sending 10% of your traffic

2 Answers2