Hi Guys,

I want to invite any hosting companies to list themselves with our hosting reliability robot:

http://65.108.35.73/hosting

Thanks,
James D. Brausch

Do you check from multiple locations or just one? Usually networks go down and routes from one point to another gets cut.

No. We check from one location.

We remove records if a majority of hosts are down during a check period. In that case, we assume the problem is on our end.

If a majority of hosts are up during a test period, the remaining hosts that are down are reported as such regardless of where the problem actually existed. A surfer doesn't care if it was your upstream provider. They just care if they can get to your site.

James D. Brausch

Originally posted by jbrausch

If a majority of hosts are up during a test period, the remaining hosts that are down are reported as such regardless of where the problem actually existed. A surfer doesn't care if it was your upstream provider. They just care if they can get to your site.


That methodology is slightly flawed. Your provider may be having peering problems with another provider, giving you a false negative. The hosting company may be responding fine from the vast majority of the Internet, but not from your particular location.

You also don't mention how you are measuring uptime, are you using ping or HTTP get requests? An HTTP get request is going to be more forgiving than a ping, as it has a longer timeout period.

I think your service would be better served with full disclosure. IOW let people know what location/backbone you are testing from, and what methodology you are using (ping/http intervals between tests, etc).

One other thing: You may want to come up with an easier way for hosts to put themselves in/out of rotation. If a host is performing sheceduled maintenance on their server, they may not want to have that time counted -- especially if the majority of their sites are still going to be available.

Thanks for your feedback.

All the hosts listed are equally susceptible to "false negatives". I don't anticipate that people looking for a reliable host are actually going to use the numbers literally. They will use the numbers in comparison to the numbers for the other hosts in the list.

We use an HTTP GET request. It is a 1.0 request and a 200 response is considered "up". Any other response or no response at all is considered a "down".

Testing intervals will change over time. They are rather rapid currently (5 minutes) in order to gather enough data to be statistically valid for the early hosts being tested. I imagine we will be trottling back to a frequency of about hourly, but adding routines to make the actual test time random soon.

We are already planning to automate the add host/remove host routine.

I'll add that information to the page during the next robot update.

Thanks again for the feedback!

James D. Brausch

Originally posted by jbrausch

All the hosts listed are equally susceptible to "false negatives". I don't anticipate that people looking for a reliable host are actually going to use the numbers literally. They will use the numbers in comparison to the numbers for the other hosts in the list.


Two more comments:

1. All hosts are not going to be equally susceptible to false negatives. Those on the same backbone as you will be less susceptible, and if they are on the same backbone, those on the same coast as your server will be even less susceptible.

2. Remember, we are talking about people who believe that ***** is a good host because they are the "featured host", or whatever, on many hosting directories.

I'd also like to make a suggestion about your health checking methodology:

You are using a single HTTP get from a single location. I would suggest dumping the fist negative response and only counting a host as down if it fails two checks in a row. This will help account for network anamolies that show as false negatives.

Of course these are just my suggestions, and I am sure there are plenty of flaws in my logic, so take them for what they are worth.

[changed to to two -- duh, and flawed holes to just flaws]

Thanks! That's a great idea. I'll put it on the list.

James D. Brausch

I forgot to answer your first comment. Yes; that is true for now. The robot will eventually have multiple sample points to deal with this issue. The current server is in Northern California connected to the AT&T backbone, so there is little bias to any North American based hosting company.

James D. Brausch

Hi,

I have incorporated the double-test suggested above. A server will only be counted as "down" if it fails both requests.

Thanks again for the feedback!

James D. Brausch