Hacker News new | past | comments | ask | show | jobs | submit login
Shared Hosting is Doomed (and I have the graphs to prove it) (clubtroppo.com.au)
14 points by jacques_chester on July 10, 2008 | hide | past | favorite | 18 comments



As someone running a VPS host company, I disagree with the article.

It cost me far more administrative overhead than to run a shared hosting arrangement.

Firstly, although the onus is the customer to be administrator, they need plenty of help and hand-holding. Many times they are happy to pay for this help - but now we are looking at a different market completely. (The smart VPS hosters are using forums and wikis to let the customers help each other out.)

Secondly, VPS customers need far more disk space, RAM, and CPU power than shared hosting. The more infrastructure you have to maintain, the higher the complexity and administrative burden. VPS customers usually demand higher availability too.


My point is not about the current market. It's about how trends in hardware and wages mean VPSes could eventually come on top.

I've had comments from other hosts almost the opposite of yours :)


"About the charts: All charts in this post are illustrative, not data-driven."


I'd be happy to see data if anyone has any. But the whole piece is built on ratios and trends more than any particular datum.


What'd you make them with, out of curiosity?


OmniGraffle for the line charts and Numbers for the pie charts.

I've learnt through experience that people take you more seriously if you have pretty charts. Even nerds who should know better :)


Ah...Mac. Damn, I've been looking for something on Linux to make graphs like that... simple "economics style" (no actual data;-) ones with basic curves and labels.


Yeah, wish I could help. But Mac == pretty graphics is a cliche for a reason.


One thing you are forgetting is that software is also getting more complicated. Meaning you need a faster machine to run it.

If software needs are increasing at about the same rate as the hardware (and I think they are), then you are wrong.


What Intel giveth, Microsoft taketh away?

I'm not so sure. It's not as though Apache has gotten that much bigger. What's changed is the patterns of architecture; people getting lazy and then being stomped by traffic.

But you always have a kernel. You always have a web server. That part is progressively reduced by hardware advances.


Apache can get bigger though. Right now it runs in threads, and in-process php parser (for example). Why? Because it's too slow otherwise.

But get a faster machine, and people will get lazy. It's hard to code apache like that - much easier to have one process per request, and use one-run cgi scripts. Once that's fast enough, that's what people will code for.

This is just an example, but I still think applications will manage to consume any extra CPU that comes their way.

Example 2: Disk capacity, three years ago you would never have thought of a reason you'd need much storage, and now we have videos.

The faster the desktops get, the bigger (and higher quality) the videos will get. That won't stop till a single 2 minute video will take 250MB. (At that point we'll need bigger monitors.)

Example 3: Memory. One reason to use SQL cursors is so that you don't need to store the entire result set in memory at once. And used to be everyone programmed that way. But not anymore - memory is plentiful enough that queries are typically stored in their entirety in memory, and then walked through. They still use cursors to do that, but give the languages time to mature and you'll find it'll turn into random access arrays.

It's the nature of people - they won't spend time to optimize something unless they need to.

All your arguments assume usage (CPU, disk, bandwidth, memory), holds steady. Now assume they won't, assume they'll go up, and re-run your arguments.


Good point and well put. I'll give it some more thought, but my initial hunch is that the ratios of overhead will still converge, if for no other reason that software which doesn't expand as quickly (the kernel) gets factored out by the hardware-driven convergence on zero.


... Graphs you pulled out of your ass.


There is no requirement in the Graphing Bylaws that states graphs must come from real data. They are just another way of graphically displaying what he believes to be true. Feel free to argue with his beliefs.

Blogs are about conveying opinions. I don't see the problem with using a graph to convey an opinion graphically.


When someone claims they have the graphs to prove something, it is implied that they are backed up by something other than conjecture.


I'm sorry but that's nonsense. People draw graphs all the times to prove things w/o using data. Especially when trends are what is important to convey in a graphical sense.

It's not any different then supply/demand curves for different types of goods in economics. The graphs represent a concise way of stating assumptions upon which the argument is based.


I interpreted the "I have the graphs to prove it" as humor. Obviously the graphs prove nothing, but I found they helped explain the arguments.


Which is why there's a disclaimer right at the very top that they're illustrative.

If you've got data going to any of my arguments I'd love to see it.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: