If you're site is with all your customers sites... I.E. - either on the same server or dedicated... and the internet goes down in your area, or your server dies, or whatever - your site goes down..

How do you insure that the site doesn't go down?

I noticed in most DOMAIN control panels you have up to 4 nameservers set... would you just have a second server in a different place and set nameservers 3 and 4 to that? so if your 1st server goes down, the second one is available?

your DNS servers won't do you any good if your web server goes down. i know some of the more high bandwidth sites such as yahoo and microsoft use multiple web servers and just spread the load between them all. however, this is a very expensive way of doing it.

Well, basically, if a majority of my clients servers are down, I don't want my site to be down aswell - I want them to be able to submit TT's and email or call me...

Or just let post what the problem is on the front page so they know when they go to my site

Put your site on the best, most expensive network you can get :)

ImLagging is right. Having your nameservers on different servers is a good idea if you have multiple servers. If you only have one, it doesn't really matter as if that server is down, doesn't matter if the DNS resolves or not, the site will be down.

Depending upon your budget there are different options.

The most economical is to use a good host, good equipment on a good network. If uptime is paramount, choose one with premium bw providers and stay away from low cost providers.

The next best method is to have two or more servers and load balance them. If one goes down or needs to be rebooted, the other will pick up the traffic. This also allows you to put more sites on each server.

The best method is to load balance two or more servers in different data centers in different geographic areas. For example, have server 1 in a DC in City A and server 2 in a DC in City B. Ideally they will have at least one provider which is not in the other, for example:

City A - Chicago - BW Providers UUNet and AT&T
City B - San Diego - BW Providers UUNet and Level 3

With three different providers it is unlikely a huge outage on UUNet will affect both networks.

It's all about money and how important it is to have your sites run optimally.

Originally posted by Brad@RackMy
With three different providers it is unlikely a huge outage on UUNet will affect both networks.

It's all about money and how important it is to have your sites run optimally. [/B]

great advice....however you failed to mention one key point:

It doesn't matter if you don't know how to correctly run your network in the first place.

Having a redundant network is great but you have to consider how your authentication (shell,mail,etc) systems are designed, you also have to consider how your data will be replicated to the other redundant host(s). When a user ftp's a file onto one of your servers, how does it get copied to the other machines that are serving the same site?

For that matter, how are you going to balance the traffic? Are you going to use round robin DNS? Are you going to use a hardware load balancer?

etc...It requires a LOT of planning and obviously the experience of having done it before.

I don't know a whole hell of a lot about the market but I feel fairly certain that this issue is going to be one of the main catalysts to the weeding-out of web hosting businesses. *shrug*

Originally posted by scott77


It doesn't matter if you don't know how to correctly run your network in the first place.


Guess I shouldn't have assumed the host would know how to do this. Any host offering this type of load balancing should have engineers on staff who do know what they are doing!




Having a redundant network is great but you have to consider how your authentication (shell,mail,etc) systems are designed, you also have to consider how your data will be replicated to the other redundant host(s). When a user ftp's a file onto one of your servers, how does it get copied to the other machines that are serving the same site?

For that matter, how are you going to balance the traffic? Are you going to use round robin DNS? Are you going to use a hardware load balancer?



I would recommend using a hardware solution vs DNS. DNS is less expensive but not as reliable. If everything is set up properly, changes should replicate to the other server automatically. Usually it takes a little time (I think the time can be set) for the replication to happen (not real time), although that is a question for an engineer. If you want me to connect you with one of our engineers, pm me.

Zoneedit.com has a "Failover Service", has anyone tried it?
If your site is down, it's supposed to send you to another.

so, if you duplicate your site onto two server's, your site should always be up. Is this what you want?

I don't know about your db's though, lol

And can't you have backup mail server's?

Originally posted by Brad@RackMy


time), although that is a question for an engineer. If you want me to connect you with one of our engineers, pm me.

I am a UNIX sysadmin, don't need any help thanks ;)

Originally posted by mattschinkel
Zoneedit.com has a "Failover Service", has anyone tried it?
If your site is down, it's supposed to send you to another.

so, if you duplicate your site onto two server's, your site should always be up. Is this what you want?

I don't know about your db's though, lol

And can't you have backup mail server's?

You can do this yourself simply with round robin DNS (multiple A records for one hostname....check out nslookup for www.yahoo.com some time)

Database servers, yes you generally want to replicate them with their built in replication functionality. Which means you're looking at a commercial SQL server most likely (Sybase, Oracle, ...)

mail servers - yes you can have multiple mail servers by using multiple MX records but the lower priority mail servers won't actually DELIVER mail to mailboxes. They usually just hold it in their queue until your main mail server comes back online.

Now.. I have a question

could you .. lets say

Have a server at datacenter A with nameservers ns1.domain.com and ns2.domain.com, let it propogate, then make a server at datacenter B with nameservers ns1.domain.com and ns2.domain.com (same exact nameservers from the 1st server) and let it propogate.. then unplug it or something..

using server from Datacenter A as your main site.. if something happens, you have the server from datacenter B with your emergeny backup... so that way, if something happens to server in datacenter A, you just plug in server at datacenter B, and all is well...


Would that work?

Nope... that wouldn't work. First of all... if I am understanding you correctly... you can only have one active IP for each of ns1.domain.com and ns2.domain.com. So... once you make the nameservers at datacenter B... the nameservers at datacenter A would no longer be propogated.

Sort of a variant on the above... you could have ns1&ns2 at DCA and ns3&ns4 at DCB. Then, if your sites fail at DCA, you can change the DNS entries at DCB to reflect a local webserver there. Unfortunately this will still cause downtime for your sites and the switch will not be instant. (If you have the IP for the site cached by IE... it could be 4 hours before the DCB server appears to be active).

Also, with round robin DNS... there can be complications. For example... apache crashes on the server at DCA. The nameservers at that datacenter are still up and sending requests to the webserver there, which will cause downtime. However.. as long as you monitor your services, the downtime will be minimal (for example... oops httpd is down... login and restart it.. voila! fixed!) Also - if someone is browsing your site and happens to be directed to DCA by the RRDNS... if DCA goes down, this visitor will not be redirected to DCB instantaneously, and your site will appear to be down.

I think round robin dns is more hassle than it is worth if you are using it for redundancy. I'd say just make sure your main site is on a good network and high quality server. If the network goes down, chances are it will not be down for long. If the server goes down, you could always have a hot standby server (maybe even a cheap-o one). Then just call up the dc, have them swap the ethernet cable... flush the arp cache and you're back up.

(Sorry for the rambling in this post... I just kinda got rolling :D)

Replication doesn't necessarily mean a commercial OS. Even relatively simple replication offered with MySQL can be fine. We replicate across multiple servers dotted around europe, and when one shut down because the CPU core hit 103 degrees after the CPU fan died, the database and infrastructure design meant that our standby servers took over and we lost no data etc., and only had fairly brief unavailability.

The technical advice on this thread is excellent, I learned a little...

Originally posted by Lirath
Now.. I have a question

could you .. lets say
...
Would that work?

Kinda. The theoretical way to do this is by getting a decent-sized address block from ARIN directly instead of your colo provider. Once you do that, you talk your colo provider & everyone into routing your block throughout the Internet.

Then you get several different colo facilities across the world and at each one get a direct point to point T1 connecting them all in a circle.

Split your block into subnets, one for each colo facility.

Then set up DNS servers at each one, where each one thinks it's primary for the domain, except the IPs that it resolves www.yoursite.com to are in its local block. Also each DNS zone will list your other name servers as SECONDARY nameservers. NOTE: you DON'T want to set any nameservers up as slaves, just add them as NS records in your zone.

So in essense you have N dns servers answering for yoursite.com to regionally local traffic. So for instance Asia theoretically gets routed to your datacenter in Asia, US gets routed to datacetner in US, etc.

Then you replicate all your data throughout them.

Then you jump for joy because you just facilitated a theoretically 100% redundant network design! Woohoo!

Good luck!

(disclaimer: I just pulled this out of my ass, but this is the general idea. don't expect to even get past step 1.)

Originally posted by mpope

Also, with round robin DNS... there can be complications. For example... apache crashes on the server at DCA.

Well I never claimed it was a solution without faults.

I am just starting my hosting business, and my 6-12 month plans are to go from one machine doing everything, to a back end fileserver (probably also handling mail, depending on the financial situation) and multiple web servers NFS-mounting the data from the file server. the web servers will probably be balanced via round robin DNS but it is fine for me because they will be in the same facility and I know what I'm doing, so...

This will allow me to basically scale to N web servers and once my fileserver is at capacity, I buy another one and either distribute clients across the two (maybe do it by username or site name in alphabetical order or something) or just start adding new clients to the new machine.

It allows for a bit of redundancy without spending a metric assload, that is my goal. And now I just told the world so everyone can steal my design!! haha

I always liked this thread

http://www.webhostingtalk.com/showthread.php?s=&threadid=74424&perpage=15&highlight=RAID&pagenumber=1

but I think you need to buy hardware to do clustered hosting.

scott77,

Have you looked at the H-sphere control panel? I don't think it does the redundancy quite (i'm sure you could probably implement it), but it is excellent for Nix/2k hosting in cluster style environments. You can setup your servers so that you have like (depending on your finance)

Unix 1 DNS/Mail
Unix 2 DNS/WWW
Win 1 WWW

I like the H-sphere system because it doesn't put too much stress on the Windows server, as iir it only runs webserver and DNS/Mail are handled off-server allowing for less things to go wrong imo.

URL - http://psoft.net if that's any help to you.