Sunday, 6 July 2008

Geographical Computer Networks

Today we are going to talk about geographical computer networks; these are among some of the most complex networks however the underlying structure is the same as any other network.

There are two network models Centralized and Regional sites.

We are going to cover QoS, Proxy’s and Round Robin DNS, most of the information needed to so this I have covered in previous postings.

Centralized has an ease of management and backup but is often slower for users then Regional sites and in addition it puts an overhead on WAN links as all requests are traveling via the WAN and this also leads to a single point of failure if the WAN link stops working.

Redundancy: Server are in a highly stable environment. However they are dependent on WAN links to users.
Availability: Using clusters and Load Balancing 99.99% up time can be guaranteed
Performance: Is good for local site or sites with strong links but many Regional site will perform poorly.

Regional Sites make use of local resources but cost more in administration time because of the complexity of the solution, however but making use of Round Robin DNS to find the closed resource.

Redundancy: Not dependent on local hardware as traffic can be moved to WAN link should local hardware fail. However larger number of servers needed.
Availability: Using nested named resource gives 99.97% uptime with always on resource being online.
Performance: Is good for users as load is balanced between resources both local and remote.

WAN Network optimization
WAN links between sites can become loaded with a large number of unneeded packets most common among these are NetBIOS broadcasts, UDP packets for this reason you should be sure of the traffic that is need and passing over your WAN links

By using WINS and blocking broadcast traffic on your routers you can reduce the UDP packet load however services and applications that use UDP normally such as VoIP and Streaming applications can be protected on the LAN with QoS and TCP trunk being sent over the WAN a good example of this can be found on Cisco site

SQL Geo Network
One of the biggest problems is Microsoft SQL server as yet there is still no geographical solution for data replication the closest at the moment is Microsoft SQL 2005 Merge replication, this can be used to create fault tolerant solution where by each production server hosts a read/write copy of the database. The databases are kept in sync using SQL Server 2005 peer-to-peer replication. Applications connect to the SQL cluster through the production interface using a host name that will distribute and load balance traffic between the nodes.

Redundancy: There is a copy of the database on each of the different servers. If one node of the cluster becomes unavailable, the other nodes automatically pick up the traffic.
Availability: Each server can be taken out of the cluster individually so maintenance can be performed without causing the database to become unavailable.
Performance: Application calls to a database are load balanced between the four nodes of the cluster. Balancing the load should result in better performance during times of increased activity.
Ideally you should have two servers in large Regional Sites so that if there is a fault on one the load is not sent over the WAN to the next nearest site unless needed as the WAN link is primarily used for replication traffic.

Round robin DNS
Is often used for balancing the load of geographically-distributed Web servers. For example, a company has one domain name and three identical web sites residing on three servers with three different IP addresses. When one user accesses the home page it will be sent to the first IP address. The second user who accesses the home page will be sent to the next IP address, and the third user will be sent to the third IP address. In each case, once the IP address is given out, it goes to the end of the list. The fourth user, therefore, will be sent to the first IP address, and so forth.

In windows 2003 Round robin DNS such can be used to matching the request to nearest subnet so that traffic remains with local resources first.
However word of warning this will place more load on your DNS servers so make use you have enough of the to take the load.

Hardware Consolidation
The Consideration is the amount of hardware and cost centralized solutions have fewer server and therefore lower administration costs but the hardware is often costs more as I higher fault tolerances is needed.

The fewer the number of servers the lower the administration cost so in this example I have used only 3 physical servers as the Active Directory server Web and ISA server are all virtual servers while the SQL servers are two physical servers.

Personally I think that you can use one blade centre to build a complete site however since you need a SAN for large storage the cost is more than most want to pay for a solution.

No comments: