Server type - does it matter?

Sparked by the 'server statistics' post ... I didn't want to hijack that thread, hence this one.

In all seriousness (and I'm really clueless about this stuff) - having been moved to an eval server for excessive CPU use I do wonder a bit. How dependent on the nature of the server itself is the assessment of what is 'more than your fair share' of CPU use? Is there some fixed number of seconds you get to use, regardless of what % of total availability that amounts to? How do I know what is a reasonable amount of CPU to consume? Is my gut feeling that the server matters really just one of those things that seems like it OUGHT to be true, but isn't?

Similarly on the issue of performance, I imagine that one person hosted on a PIII may see snappier performance than 1,000 people hosted on a quad Xeon. Is there some level of traffic where I'd be considered an OK neighbor on one server, but a greedy process pig on another? Analog says I get roughly 10,000 successful page requests a day, with roughly 60,000 server requests. Is that light, medium, or heavy traffic?

While we do have some older hardware, pretty much all of our machines (and all the machines hosting people who have signed up in the last year) are relatively similar Dual Xeons with 2GB or so of RAM. They're pretty kick-ass.

And we're not just looking at the percentage...it's also the raw number of CPU minutes/seconds you're consuming...

And don't worry, it's not so bad being on the eval server--we just want to make sure as few people as possible are affected when a site is using tons of resources!


nate.

> While we do have some older hardware, pretty much all of our machines (and all the machines hosting people who have signed up in the last year) are relatively similar Dual Xeons with 2GB or so of RAM. They're pretty kick-ass.

Sorry if it is me who hijacks this thread, but is it just me who think that long time customers are penalized here. Most people should be happy with Dual Xeon servers, but then how old are the older machines, on which the early settlers are resided? P3? What is the last time they were upgraded, or is there any plan (timetable) to upgrade them (or is that necessary, and how urgent that is, etc.)? Thanks.

Giving DH the benefit of the doubt, they probably have a carefully crafted equation similar to
N=bogomips/Slashdot_Probability/Rand()
to determine the number of sites allowed to share (overbook) each server.

Just like C0mc@st, eventually the discount honeymoon is over. :-) If you're having server performance problems, maybe they'll talk to you about moving. An older system could be better, if they put fewer customers on it; your chance of being taken down due to a "neighbor" on the same server would be less.

BobS

> Giving DH the benefit of the doubt, they probably have a carefully crafted
> equation similar to N=bogomips/Slashdot_Probability/Rand()
> to determine the number of sites allowed to share (overbook) each
> server.

I wouldn't say that we have an equation, per se, but it is true that we have a reasonably good idea of how many people we can host on a given piece of hardware before problems develop. A number of criteria go into this, ie. what plans people are on, how many 'CPU minutes' they use, etc. While it's not as scientific as you might think, we've gotten fairly good at it. Beefier servers host more people, so it all works out.

It is a bit cost prohibitive to replace servers on a regular basis when older servers are perfectly functional and perform well. It'd be unreasonable, I think, to expect these servers to be replaced as new hardware becomes available unless they're broken - at least, at the prices we offer.

Plus, by being "old" such servers are probably among our most stable and reliable servers. Such servers already have a reliable 'mix' of sites/customers known not to cause problems, as we long ago weeded out any troublesome accounts. In fact, I'd say that the least stable servers we have at any given point are probably new-ish ones. They get better over time as we identify problem areas and resolve them.

So, yeah, I wouldn't be so quick to jump onto new hardware if you haven't had trouble with the hardware you're on. I know that almost all of my personal sites are stored on a pretty old server, and I'm very happy with it. For quite a while, I even had sites on the very first server we ever had, and it too was pretty good (though admittedly it needed some upgrades over the years to keep up with what we put on it).

- Jeff @ DreamHost
- DH Discussion Forum Admin

Haggis,

I'm so glad you asked these questions. I've been wondering about how well the different servers hold up to traffic as well. I have eight domains hosted on one account at Dreamhost, and I expect heavy traffic on two of those domains (this expectation is based on experience as I used to work for a top 500 web company. It was hosted on 25 dedicated servers with a load balancer for the database. The site still frequently went down).

I know that my Code Warrior account isn't going to get me far, but until my sites start making money, it's the best I can do--but a huge worry. Right now, my websites are already
downloading much slower and sluggishly, and I've only just
begun.

I wrote to support and asked what they suggested--whether it would be best to split the domains up into two different accounts to lighten the server load or what, and I didn't receive a satisfactory answer.

They told me how to set up a second account, and move some of the domains but didn't tell me if it would be the best thing to do or how it would affect my present domains, some of which have databases.

Has anyone here ever dealt with a problem like this? I'm really not sure what to do at this time. :-)


Bobette Bryan