[wp-hackers] Forum post: Huge server hits

[David Chait]

(I wrote this before Jason's latest 'data points', which continue to show 
real-word costs and numbers...)

"Might" is because it depends on their host, their hosting plan, and their 
hosting server(s).

The problem is that many HOSTS can't even answer the question!  In some 
cases, the hosts are resellers, just making a buck.  In some they might have 
decent linux expertise, but they have no idea what it'd take to handle a 
large-scale site.

The question they could ask is: do you have a hosting plan and server 
resources that can handle a 100K/day PHP based website?  If they need a 
starting base, many of them might be familiar with one of the PHP BB 
systems, or PHPN*ke, etc., as a (heavier-weight) comparative.

I guess the answer is we could come up with some example numbers (Jason gave 
some interesting ones), we could set up a test server to bang on, but the 
reality is that when you hit 100K, 1M, or 10M, you are talking a scale 
that's beyond whether >WP< can handle it, it is whether the INFRASTRUCTURE 
can. ;)

Someone check my math:
100K/day is approx 1.16 hits per second (and by the way, 'hits' could be 
interpreted as users, as pageviews, as...).
Take a website that's a maybe 160K in exchanges for page with cool graphics, 
core html/js, and CSS.
That's about 190K/s in throughput.  Which is about 1.5Mb/s sustained 
throughput -- what many basic 'unmetered' dedicated servers are CAPPED at, 
btw.  But that's ONLY if people visit in nice succession.
So, for 10 people to hit the site simultaneously per second, sustained over 
some period, you'd need something around 15Mb/s connection (note, that's 
higher than a 10Mb connection would sustain already... I know, should be 
obvious 15 > 10, but wanted to be clear).  So, some places now offer 20Mb/s 
connections, non-sustained but in bursts.  So, you could actually burst up 
to maybe 13-14 hits per second, sustained, for some period.
Okay, up another notch.  Let's say you have a 100Mb connection.  You could 
handle up to around 70 hits per second.  That's sustaining the maximum 
throughput.
Now for reality.  You never get sustained maximum throughputs -- at least 
not without paying thousands of dollars a month for it.  The net gets 
congested, traffics gets slowed.
Second reality.  10 hits per second isn't 10 hits per second -- not really. 
There's php executing per load, retrieving the html for the page.  Then the 
browser grabs the css file(s).  Then all the images.  And much of that is 
simultaneous requests!  So depending on page content, you could be getting 
100 http requests per second easily.  Someone with apache knowledge can 
comment on that...
And that was just PURELY NETWORK.  The 20Mb/s, which was 13 hits/s, for 
let's say 15 queries, is 195 queries/s.  Someone who knows mySql can give an 
aspect of what load that'll put on a server. ;)

And as Jason pointed out, that's assuming there's no video, no podcasts, no 
downloads, just the base page.  Jason's nice to not be charging for overage 
on the bandwidth -- I don't know many other hosts out there that wouldn't 
sock you with a potential couple-grand-plus bill.  some charge like $1/GB 
overage, and if you have 1000GB extra over a base of maybe 30-300GB, that's 
a $1K charge!  but we've been talking about cases where you could be seeing 
a jump to over 10TB with millions of hits, or larger downloads like 
podcasts.  That kind of site needs hardware, network, admins, 
software/config, and bandwidth.  Hardware isn't cheap (as Jason noted), and 
unless you can get a discount provider, even pre-buying a 100Mb line/mo 
could run thousands of dollars (or more for premium, less congested, 
redundant, networks).

Just realities to think about.

----- Original Message ----- 
From: "Podz" <podz at tamba2.org.uk>
To: <wp-hackers at lists.automattic.com>
Sent: Tuesday, July 12, 2005 3:56 PM
Subject: Re: [wp-hackers] Forum post: Huge server hits


> Someone posts to the forums and says "I'm going to create a website for 
> someone famous and I expect 100,000 hits a day. Can WP handle this ?"
>
> It would seem that the response would be "Use wp-cache, it might handle 
> it".
>
> "Might"
>
> Why should they invest time energy and money when it "might not" ?
>
> The people asking these questions may have no idea about maths / requests 
> / bursts / dedicated whatever - they are asking for help in making an 
> informed choice. Is "might" the best we can offer ?
> Would it be good to read that a big site dumped WP because - as /they/ see 
> it - WP couldn't handle the traffic ? It doesn't matter what the answer 
> really is, no-ne will hear that.
>
> I do not think it unreasonable for WP to have an answer which people could 
> be told to pass to their host. Make the contents of that answer as geeky 
> and as technical as you want because right now it is better than what we 
> have - which at best is guesswork, and at worst nothing.
>