NYCPHP Meetup

[leam]

On Sat, 2008-01-19 at 10:20 -0500, David Krings wrote:
> leam wrote:
> > Not a question of "can you make it more performant?" as that's easy to
> > answer. I'm looking at "Is there a web metric that makes it more viable
> > for a small start up to spend time maintaining non-mainstream software
> > collections than using standard tools like Apache, Linux, and
> > traditional servers"?
> 
> Do you even need these improvements? If your current setup works out, why 
> bother optimizing it and potentially hitting a bunch of problems? Improving 
> performance is always appreciated, but finding metrics, evaluating them, then 
> tweaking them, then trying different constellations. I don't know, but that 
> effort and risk for a theoretical performance increase?
> 
> Some long time ago I read in one of the more populer computer magazines (Micht 
> have been Computerworld) a performance analysis of IIS on Windows (obviously), 
> Apache on Windows, and Apache on Linux. Their results showed that Apache on 
> Windows performs best. Their recommendation was to use the WAMPP stack for web 
> development. I don't know how accurate their test was and how the situation is 
> today, but it shows that there are more factors to performance than just the 
> web server application. Sure, some light weight web server apps can surely 
> perform better, but there are likely things that these apps can't do. And 
> other characteristics such as reliability are almost more important than kick 
> ass performance.
> 
> David


Lack of performance costs enterprises billions of dollars in datacenter
footprints annually. Performance at the "5 clients on my web-host" level
is, as you note, less an issue. However, I think big and scaling is an
issue. 

Let's say I want to charge $20 a month for web-hosting, offering 50 M of
space. I just configured a Dell PowerEdge SC1435 with 8G ram, dual 300G
hard drives and two dual core Opterons. Comes to just over $3000, for
redundnacy I need two. That means a lease of ~$280/month for 36 months,
or 14 paying clients for the hardware. Server is small, so 5 clients per
month for rackspace (2x 1U@$50/U). Using Open Source software all the
way, so that's a cost savings.

However, I'm a start up so I want to grow. One tool is a re-seller
program; I'll let my ad hoc sales force take 50% of the income from
paying clients just to get growth. That means my "19 clients to pay
expenses" becomes 38. Disk space isn't an issue, but support operations
time is. If I can make $1000 a month doing side work for someone else I
need to have 100 more paying customers to make this business venture a
break even, and that's not counting administrative overhead, marketing,
insurance, etc.

So we're not at a disk issue yet, but hypothetical load issues need to
be attended. While the servers are technically 2x Dual core 2.4Ghz with
8G of RAM, they are effectively half that because I need both machines
to be able to handle the entire load. That provides for hardware
failure, power or NIC failure, etc. 

So we return to metrics. While the boxes are pretty nice, at what client
count does their performance become "slow" and is it better to upgrade
hardware or spend personell time tuning the servers? Business profit
should derive from value added to the product, what vaules are worth
adding and charging for? If there is no noticeable market response
between a 20 second web page and a 30 second one we can allocate
resources elsewhere. If there *is* a difference then knowing and tuning
to beat the metrics becomes a business advantage.

Side note. No, this isn't a "for class" assignment. I'm actually
thinking through these issues because I see some good business
opportunities here. If you are in this or a similiar business and don't
mind sharing your insights, feel free to contact me off line. Hopefully
I'll produce enough clear thought to help others down the road as well.

Leam