Ian,

Inline...

-----Original Message-----
From: Ian Griffiths [mailto:[EMAIL PROTECTED]] 
Sent: Dienstag, 7. Mai 2002 11:18
To: [EMAIL PROTECTED]
Subject: Re: [DOTNET] ASP.NET's Application object: does it scale?

According to Microsoft, the www.microsoft.com web site gets 20 million
page
views per day[1].  That's some 4 million short of your definition of
large.

*** Correct. MS website - and I assume you talk of www.microsoft.com -
is also very static. From the web traffic site this is NOT a large site,
sorry. What makes MS large is the amount of content (80 GB) and the
point that they handle a lot of traffic through downloads on their site.
The web site by itself is not exactly a sample of a large etc. site and
would in itself NOT need the bandwidth the have - here we talk about
downloads.

(Remember that they have an international audience, so their usage
patterns
are not compressed into an 8 hour space.)

So you're saying that www.microsoft.com is not a large web site?

*** Not from a traffic point of view.

In any case I think your reasoning here is suspect:

*** No, MS is not a regular site - they are by large amounts a download
center and have a tremendous amount of content, writers etc. BUT the
pure web serving is NOT ridiculously high. I know a lot of sites that
have higher traffic in terms of pages served.

>1 million page views a day, lets say over a period of 8 hours
> (to account for spikes etc.) are 125000 page views an hour.

Judging by every web site and internet link usage graph I've ever looked
at,
you don't get uniform distributions like this.  I understand that you've
compressed the day's traffic into working hours, which will give a
higher
'average hits per second' figure than averaging it out over the whole
day,

*** This was my idea - actually I was getting a 3x peak.

but I think you will still be underestimating the peakiness of the load
-
you usually see spikes in the morning and at lunch time.  (On the other
hand

*** Hm, 3 times is still pretty high :-)

if your web site has international readership, it will be a bit more
spread
out.  You'll have more peaks of course, but they'll be less high.)

At peak times you should expect well over 35 page views a second.
(Well,
assuming your server can cope with the load...  A lot of servers slow
down
at lunch time.)

I tend to think of web sites as being (very approximately) one of:

  (1) Dead (e.g. no page views most days)
  (2) Piffling (a few, maybe a few hundred)
  (3) Light load (a few thousand)
  (4) Non-trivial load
  (5) Large
  (6) Huge

(1) through (3) would not be worth putting on their own server.  (E.g.
most
personal web sites, and most small company web sites.)  My criteria for
moving into (5) is approximately when you really need more than one
server
to handle the load.  (And I mean *need*, as opposed to using multiple
servers because it's cheaper than making one server work properly.  It's
surprising how much bandwidth a single server can chuck out if the site
is
designed well.)  I'd say that a million hits a day was easily in
category

*** FULL agreement on this. Actually you go into more servers often in
order to
* keep your SQL Server "in the back" or
* actually handle processing like in the SQL Server.

(5).  A million an hour is definitely (6), but how many of those are
there?
Not even Microsoft get a million page views an hour on average.

*** A million an hour is NOT 5 - it is 4 or 5, but mostly 4. Let me get
some little numbers straight.

A sample of 6: YAHOO:
1.32 billion page views per day on average during December 2001

Get this - that is 1320 MILLION PAGE VIEWS PER DAY. That is roughly 50
million an hour. THAT is huge.
Reference: http://docs.yahoo.com/info/pr/investor_metrics.html

Check the access numbers of hotmail, amazon, ebay, and there you have
sites of 5 or 6. I consider everything below one million an hour to be
maximum 4, UNLESS the processing is extreme.

Regards

Thomas

You can read messages from the DOTNET archive, unsubscribe from DOTNET, or
subscribe to other DevelopMentor lists at http://discuss.develop.com.

Reply via email to