Hi

I see 99% and 99.9% up-time offered at web hosting sites.
Could 100% be reached, here I am thinking of hosting my sites on several servers at diferent locations. I think the minimum would be 3. If any one server went down then the other two would still be live.

A reseller could open three accounts with diferent hosts, and place any 100% up-time web sites on all three.

Could this be done ?
How?
Has anyone done so allready?

Some sites I hope would pay more for this service if it could be done

Regards
Michael

if you want to pay triple the cost for .1% higher uptime, that is crazy unless you are bringing in a ton of money. if you sign up with 1 host most of them have 2 servers they put you on (ie ns1.bluecityhosts.net, ns2.bluecityhosts.net). I have never seen in DNS setup more than 2 slots and then an optional one. Try if you want, but don't expect people to pay a lot more for it.

Even biggies like Yahoo! and eBay can and have gone down because of DOS attacks and equipment failures. I don't think 100% under-any-circumstance uptime is possible, although 100% with-a-bit-of-luck-and-planning uptime may be.

Anyone with 100% uptime is probably lying. I know I could easily take any system down by writing a simple CGI script if I really wanted to. I got bored and did it to one of our systems one time just to see and it owned the system within 15-20 seconds with no users on it. On top of that if you ever have someone that gets pissed at a customer of yours they can just DoS the server for a few minutes / hours / day / weeks (if you last weeks you need to read up on filtering heh). But anyhow.. thats why nobody in their right mind would offer 100%

There seems to be a common misconception about uptimes. It is very well possible to offer 100% uptimes if you are connected to a redundant BGP4 backbone. This is not saying you will not have any downtime but if there is an unscheduled network outage you will be reimbursed for anything under 100%. Its the same thing with 99.999% , 99.5 etc. if network uptime dips below these percentages you will be reimbursed accordingly.

As for the original question you can do this using your method however keep in mind every account on one server located in one data center would have to be mirrored to a second and third server and datacenter. Below is a brief description.

Data Center 1 Server 1
domainname.com
ns.domainname.com (Primary DNS)
all accounts on server

Data Center 2 Server 2
mirrored domainname.com
ns2.domainname.com (Secondary DNS)
all accounts mirrored from server 1

Data Center 3 Server 3
mirrored domainame.com
ns3.domainname.com (secondary DNS)
all account mirrored from server 1

With the above configuration if data center 1's network would go down, the request would go to Data center 2 and so on..

Keep in mind that the IP addresses used will be different between Data Centers which could cause some conflicts until the original data center is back up and running.

100% uptime spreading across 3 servers.

Where did this number 3 come from?

What happens when Server 1, Server 2 and Server 3 become unaccessable at the same time?

Is the theory behind the 100% simply that it can't/won't happen?

I'm just curious where 3 came from, as opposed to 6, 13 or 100.

In the original post, he mentioned at least 3 different data centers and servers so that is the number I went with. The theory of 3 different data centers going down at the same time is unlikely, and if it did occur, everyone would probably lose internet access .

The true way of having 100% uptime is to have a data canter or data centers that use redundant BGP4 routers and true Teir 1 providers. This way if one backbone goes down it switches to the other and so on. By having redundancy this lessens the chance of the dreaded downtime, not saying it wont happen but if it does you will be reimbursed accordingly from the moment of downtime, not 20, 30, or 40 minutes afterwards depending on the SLA percentage.

Hope this clarifies things a little better from my post above.

Hi

My main idea is how close can you get to 100% up-time and would you be able to reach true 100%. That is, a web site would be avalable even when one server was of-line for repair or maintainance. If a customer came to you and asked for a 100% up-time and was willing to pay any extra!

This could be a selling point! The must be some sites that would/need this.
But you would have to show the customer how you planed to get that 100% up-time.

Why 3 servers, well one server seems to much like putting all your eggs in one basket. NS1 and NS2 in the same data center is still asking for troulble. Any more than 3 seems like over kill. If all 3 servers/data centers on three diferent areas of the world went down then WWIII may well have started!

Regards
Michael

Originally posted by Michael96
Hi

My main idea is how close can you get to 100% up-time and would you be able to reach true 100%. That is, a web site would be avalable even when one server was of-line for repair or maintainance. If a customer came to you and asked for a 100% up-time and was willing to pay any extra!

This could be a selling point! The must be some sites that would/need this.
But you would have to show the customer how you planed to get that 100% up-time.

Why 3 servers, well one server seems to much like putting all your eggs in one basket. NS1 and NS2 in the same data center is still asking for troulble. Any more than 3 seems like over kill. If all 3 servers/data centers on three diferent areas of the world went down then WWIII may well have started!

Regards
Michael

Exactly what I was stating in my posts.. :D)

If I had to offer someone a true HA solution, this is probably what I would recommend:

Big Picture:
- servers spread across multiple data centers w/ geographical load balancing system

Data Center Level:
- absolute minimum of 3 Tier 1 bandwidth providers / multiple telco fiber feeds into each data center
- hot stand-by core routers
- cross connected switches
- redundant local load balancing system
- small server farm hanging off the load balancers

And all of this would cost what? Probably $100,000 or so.

:D :D :D

I think it can be done but will cost alot of $ $ $

This means NO DOWNTIME = hmmmmmmmmmm

:confused:

I was just pointing out that 3 is a rather arbitrary number that may not be as magical as you think. I have no idea, but I'd do a little statistical math and figure it out.

If you were insuring your business based on 100% uptime, you can bet the insurance company wouldn't think, hey, 3 servers, they would never all go down at once. It's 3 servers!

But I guess I'm being picky. :)

One more thing to keep in mind, even though I did use 3 just because that was the number he stated, you also have to remember that if the data centers you choose to use have redundant BGP4 and true TEIR 1 connections this means there are normally 4+ pipes from different carries in each data center, so if we use the example of 3 data centers this is in theory, 12 different pipes available for use, and as stated above if all 12 of these went down at the same time the whole internet would more than likely be down as well.

3 is not the magical number, in fact I really dont think there is one, it really depends on who you choose for a co-location provider and how they are connected.

...if the data centers you choose to use have redundant BGP4 and true TEIR 1 connections this means there are normally 4+ pipes from different carries in each data center, so if we use the example of 3 data centers this is in theory, 12 different pipes available for use, and as stated above if all 12 of these went down at the same time the whole internet would more than likely be down as well.

Thats all great, but what happens when the ethernet cable connecting your machine gets caught on the sleeve of a tech and gets disconnected? Or an electrician drops a wrench into a circut box? Or a plane crashes into your NOC?

There's a lot of things that bring servers off the internet.

Originally posted by driverdave


Thats all great, but what happens when the ethernet cable connecting your machine gets caught on the sleeve of a tech and gets disconnected? Or an electrician drops a wrench into a circut box? Or a plane crashes into your NOC?

There's a lot of things that bring servers off the internet.

If these type of things all happen at the same time, at different locations, at different NOCs, then I would have to crown you the worlds most unluckiest person, or the most paranoid.

Or a plane crashes into your NOC?Wow, I have never see that :) I do know a couple of datacenters that if the NOC were hit by a plane, the datacenter would still function as they are isolated.

I think the first step towards this is SLA's from every provider that provides services that make your webhosting company run. Not just network, hardware, electricity, phones (100% uptime support) aircon etc.
Now I dont know about the US but here in OZ a 100% SLA for just about anything is near impossible to get.

Originally posted by RackMy.com
Wow, I have never see that :) I do know a couple of datacenters that if the NOC were hit by a plane, the datacenter would still function as they are isolated.

Mike, I was hoping you would show up at this thread eventually. :D

As you can see this is a host that does provide a 100% uptime SLA through a redundant BGP4 data center.