Web Hosting Talk : Web Hosting Main Forums : Hosting Security and Technology : Complex Mirroring project , need suggestions !

[Allwin A]

Hello Folks,

We have an interesting project to work on. A client has issued a clear directive to us. He wants his server up @ all costs. He is willing to have his server mirrored at three DCs in different time zones. He just wants his server up serving data even if any of the other servers are yanked off.

The OS has to be Windows and SQL too is necessary. How can all of this be mirrored across servers in different DCs and most importantly how can each of them go alive if one of them goes down. Please let me know what kind of options are available if we intend to pull this off

Cheers,

[pnorilsk]

Actually you don't have many options, IMHO. The selection of Windows put some serious constraints, but it's still doable. The database selection and implementation could be a major issue.

Peter Kinev.

[eth00]

Having all the servers in different datacenters make it hard to have true 100% uptime. If one goes down you are going to have to update dns which is going to take time and that is assuming that isps do not cache the ips making the website appear now.

IMHO you would be better off hosting them all in a world class datacenter with 100% network uptime guarantee, rackspace comes to mind, then use a load balancing solution that will detect any problems instantly.

As far as the window side...no clue I stick with linux for servers

[apollo]

if you have many writes to the database... that (database replicaton) could be a big problem....

[dollar]

i second the call for rackspace

[Babushka99]

There are two ways of approaching this problem. One is to have a Primary Clustered (1+1) environment, this is where the "source" servers would be, i.e. your Web Server & MS SQL Server.

Make sure the DC where the Source is located has multiple-bandwdith providers and paths to it.

Next, you simple sign an agreement with any CDN (Content Delivery Network) company, i.e. like Akamai, Speedera and many others.

What they will do is (for example) provide you with POPS (reverse proxy content servers) in three locations - lets say LA, Chicago and NYC.

All the three locations independently connect to your source and push content out to the consumers. No one other than the 3 POPs can access your Primary Server (this is done either via ACLs on the source server network or by virtue of VPN conncectivity between the POPs and the Source-Server.

All the three POPs are physically separate networks, so even if one goes down, the other two are still available to serve content, even if two of them go down, you still have one of them to serve the content. WIth a redundant clustered environment on the source, you can be rest assured of reaching "almost" 100% uptime.

Your second approach is to have three sets of clusters with redundant architecture in three different cities. Use a Grographically Load Balanced switch like Foundry to direct requests based on DNS looks ups to the nearest location, or do load-balancing based on weigtage, i.e. East cost get 30% hits, West Coast gets 30% hits and Central gets 40% hits.

The key here would be a synchronization server. EIther one of the locations can be a Source server for the other two servers to sync up with, or you can have a fourth (external) source server. The replication delay will be there, it kinda slows your system down, but depends how much replication is being done. In a 2-way (mirrored) replication environment, with 3 geographically dispered POPs, you will get get decent performance.

Hope that helps.

Babs.

[Allwin A]

Thanks for your support folks. The client insists that the DC be situated outside the US and he still wants 2 to 3 DCs involved so that even if any of the servers/networks go down, the site remains accessible.

Babushka99, thanks for an AWESOME explanation of the possibilites

Meanwhile, lets assume, we keep 2 RAID SCSI servers in 2 locations and mirror it regularly, how can they be set that if one goes down, the other becomes accessible ? I am worried from the point of DNS. Wouldn't there be some downtime involved at least ? Or is there an advanced way of working this out ?

[Babushka99]

You need a DNS Switch. If you have two separate Servers Clusters in two separate gepgraphical locations, your DNS SWtich (read: Load Balancer) will keep a check on both the server clusters being alive or not. If one goes down, the switch will not redirect any requests to that cluster until and unless it comes back online. As far as your clients are concerned, they would not even know - as they would simply be getting response to their requests. When the failed cluster comes back up online - requests can go to it - once deemded fit by the Load Balance.

The load balancer can be setup in "either" of the locations or in a third - neutral location, totally up to you.

There is NO down-time with respect to the DNS (until and unless all your server clusters go down). In that case you are fubar'd.

You may want to read up on F5's Bip IP Global 3-DNS Controller at:

http://www.f5networks.com/f5products...cts/bigip/gtm/

Do read up on the FAQ section provided in the link above.

Hope that helps.

Babs.

Just to add above - in the above scenario - you would NOT necessarily need to employ the services of a CDN carrier/company.