[NCLUG] [Poll] VPS and web mangement hosting

Sean Reifschneider jafo at tummy.com
Mon Feb 20 13:18:27 MST 2012


On 02/19/2012 04:28 PM, Bob Proulx wrote:
> either.  I personally know one business that was spending $650/month
> for a 512M virtual server out on the net.  Wow!  I know that we all

The thing to consider there is: apparently the person with the checkbook
felt they were getting good value at that price-point.

For example, we had one customer who felt they needed a high availability
cluster, because if their ordering site was down, their customers were
going to order from one of their competitors.  Another customer had gone to
one of the big hosting providers, gotten a quote for a single server, and
then come to us and asked us to do a bunch of add-ons to get the price up
to that same level.

It's all about providing what the customer perceives as good value to meet
their needs.

> That that is easy to do.  And I think everyone takes a try it and see
> attitude.  If they can change plans relatively easily then they will

And that's very cool, especially with virtual hosting where the size can
often be scaled up quite easily.  However, it's not so cool if they stick
with their initial estimation that "this should be all I need" and are
unhappy because their "site is down again".

Luckily, we usually see this from the outside, because we can fairly easily
detect when a system is having issues because it ran out of memory and make
the recommendation to either reduce the footprint or increase the size.  So
usually it's someone complaining about another service rather than about
our own.  :-)

> Bandwidth is where I end up being completely lost at estimating.  I
> usually have no idea what will be needed.  I have no other suggestion
> than to test drive it for a while and see.  The answer is always one
> of "It depends."  I have no idea.

Indeed, and it really depends on the actual use...  In our facility, we
don't have access to the ridiculously inexpensive bandwidth so our numbers
of included bandwidth are fairly low compared to the providers in, say,
Florida or Dallas or CA.  But it's pretty high quality bandwidth, which
doesn't nicely fit into the "X GB RAM, Y GB disc, Z GB of bandwidth"
overview.

As an example, one of our guys did some testing last year and found that
his latency was often lower if he passed traffic through our VPN server at
the facility, going Comcast to our facility and then our facility to the
destination, rather than using Comcast to go directly to the destination.
Adding the extra set of hops was more than covered by our network
optimizing the path to the destination.

Not all network connections are created equal.  :-)

> My first thought was "Ha!, I always run Munin."  :-)  But you still
> don't know until it has been running for a month and has some history
> behind it.  Some tasks don't happen every day.

Indeed, and that really doesn't help deal with situations where you show up
on /. or the like...  Once a year events.

Those are places where deploying to something like Google App Engine can
really help, because it just automatically scales.  However, you have to
develop specifically for GAE.  Amazon promises similar scalability, and you
can use a more general stack for that, but you have to spend a lot of
effort in making sure that you have made your entire infrastructure
scalable.  You have to be very careful about understanding the full Amazon
infrastructure before deploying services there, unless you can afford to
lose everything...  They are not a VPS provider, they are an infrastructure
provider.

> Actually Ubuntu (and Debian too) defaults to 150 connections.  Or at

Yeah, that sounds right, I knew it was somewhere in the 100s.

> started killing things.  Linux memory overcommit is probably good for
> laptops but is definitely bad for servers.  They ended up reducing
> MaxClients to 12 down from 150.

Agreed.  You should never have the opportunity to swap on a server.

Sean



More information about the NCLUG mailing list