Systems Migration Work 2006/2007

August 2006 saw the start of a plan to implement a completely new setup at the data centre in California.

The client moves that formed the major part of this work were completed 29 March 2007.

The notes here are to inform and aid understanding regarding what this change means.

Progress Report

Last updated 30 March 2007

From the start, work progressed at a pace and the new infrastructure has now been in place since around the middle of September 2006 with the first migrations following shortly afterwards.

All Windows based hosting servers were migrated by 21 September 2006.

All vserver hosting was migrated by 21 November 2006.

All shared hosting clients web sites and email migrated by 29 March 2007.

29 March 2007: Update: All users successfully migrated on to the new systems. EcologicalHosting.com web site control panel links updated to reflect this (i.e. links to old services removed). Name server ns1 will be migrated shortly to complete the process.

What does this mean for my web site and email?

General web site migrations were seamless and didn’t cause any problems or require any action by the owner.

Changes to email setup were kept to a minimum. After migration the only change is to the POP3 username: replace the old firstname_surname format to the new mailbox@mydomain.blah format. For all email migrations, an explanatory email was sent out to the site owner at the point of migration.

Why all the changes?

The new setup creates a massive saving in power consumption, noise and other environmental factors through the use of virtual server technology (note: this is not the same technology used for our Linux vserver product).

Massively powerful new computer systems have been installed, along with huge RAID protected SANs (storage area networks). In combination with the SAN and specialist software, each of the new physical servers is capable of running multiple virtual servers through the use of machine hardware emulation. These measures result in increased efficiency giving the ability to make more use of the resources available and generate considerable savings in energy consumption as well as reducing the overall amount of equipment required.

The network infrastructure has also been uprated and is fully redundant with automated failover and full BGP (border gateway protocol) routing across three Internet service links (two wired and one very high speed wireless).

In short, this ensures even higher availability than before.

The failover and load balancing doesn’t stop with the network either. If a physical server is struggling under the load or in the unlikely event of major hardware failure, virtual servers hosted by the failed physical server will automatically be switched to run on a different physical host. This is made possible by having all the server information stored away from the physical host in the SAN storage arrays.

The SAN also brings additional benefits: Multiple snapshots are taken of each virtual server stored on the SAN throughout the day. This enables server restoration to an earlier time almost instantaneously if needed. These snapshots are created by the SAN hardware and are completely independent of the servers.

Also, from our perspective, the changes bring enhanced management features allowing even more remote control over our servers than before, including power cycling and server resource usage monitoring capabilities in addition to remote console access.

In addition, all servers continue to be monitored every 60 seconds as before.

Technical Network Overview

Below is a basic network diagram showing how the new data centre network is laid out. The explanations that follow should be read in conjunction with the diagram.

 

There are three separate Internet backbones running BGP, which route all traffic coming in and out of network via the shortest possible path. BGP also allows complete redundancy, if one or two Internet backbone links goes down, the other(s) will continue to handle the traffic.

The three separate Internet connections then connect to two redundant Cisco 7200 series routers, which use HSRP (Hot Spare Router Protocol) allowing one to take over the other during a failure. From there, each one goes to separate trunked switches, which allows one to go down and the other will take over.

Two separate Cisco PIX firewalls operated by the data centre block all but needed ports and monitor each other, so if one goes down the other takes over. Out from the firewalls the traffic goes to another set of separate trunked switches, which can failover from one to the other if one fails.

At this point the servers connect to these switches as follows: Each server has two dual port NICs (network interface cards). NIC 1 Ethernet port 1 is teamed for failover with NIC 2 Ethernet port 1. And NIC 1 Ethernet port 2 is teamed for failover with NIC 2 Ethernet port 2. Then NIC 1 Ethernet port 1 and NIC 2 Ethernet port 2 are connected to Switch A-M, and NIC 1 Ethernet port 2 and NIC 2 Ethernet port 1 are connected to Switch B-M. Each of these two port separate NIC card teams are then teamed together for complete redundancy.

All servers are clustered together and each server has two separate SAN HBA (host bus adapter) cards that give the server redundant conductivity to the SAN. The SAN is a high-speed sub-network of shared storage devices. These storage devices are machines that contain nothing but RAID hard disks for storing data. The SAN’s architecture works in a way that makes all storage devices available to all of the clustered servers. Because the data does not reside directly on any of the clustered servers, any server can go down and the other servers in the cluster will take over and balance the load.

Comments are closed.