Last week, I attended the VMware Beyond Boundaries event in London – the first of a series of events intended to highlight the benefits that server virtualisation technology has to offer. The day included a number of high-quality presentations; however there were two that had particular meaning because they were given by VMware customers – no marketing fluff, just “this is what we found when we implemented VMware in our organisations”. Those organisations weren’t small either – BT (a leading provider of communications solutions serving customers throughout the world) and Nationwide Building Society (the UK’s largest building society – a members-owned financial services provider with mutual status).
Michael Crader, BT’s Head of Development and Test Infrastructure/Head of Windows Migration talked about the BT datacentre server consolidation and virtualisation strategy. This was a particularly interesting presentation because it showed just how dramatic the savings could be from implementing the technology.
BT’s initial virtualisation project was concerned with test and development facilities and used HP ProLiant DL585 servers with 16-48GB RAM, attached to NetApp primary (24TB) and nearline (32TB) storage with each virtual machine on its own LUN. SnapMirror technology allows backing up a virtual machine in 2-3 seconds, facilitating the removal of two roles whereby two staff were solely responsible for loading tapes (with 96 hour backups of the test infrastructure).
The virtualisation of the test and development facilities was so successful that BT moved on to file and print, and then to production sites, where BT are part way through consolidating 1503 WIntel servers in 5 datacentres to three virtual infrastuctures, aiming for:
- Clearance of rack space (15:1 server consolidation ratio).
- Reduction in power/heat requirements.
- Virtualised servers and storage.
- Rapid deployment.
- True capacity on demand.
The supporting hardware is still from HP, using AMD Opteron CPUs but this time BT are using (in each datacentre) 36 HP ProLiant BL45 blade servers for hosting virtual machines, each with 32GB RAM, 3 HP ProLiant DL385 servers for management of the infrastructure (VirtualCenter, Microsoft SQL Server and PlateSpin PowerConvert), 4 fibre channel switches and an HP XP12000 SAN – that’s just 10 racks of equipment per datacentre.
This consolidation will eventually allow BT to:
- Reduce 375 racks of equipment to 30.
- Reduce power consumption from approximately 700W per server to around 47W, saving approximately Â£750,000 a year.
- Consolidate 4509 network connections (3 per server) to 504.
- Remove all direct attached storage.
At the time of writing, the project has recovered 419 servers, 792 network ports, 58 racks, used 12TB of SAN storage, saved 250KW of power, 800,000 BTU/hour of heat and removed 75 tonnes of redundant equipment – that’s already massive financial savings, management efficiencies, and that reduction in heat and power is good for the environment too!
Michael Crader also outlined what doesn’t work for virtualisation (on ESX Server 2.5.x):
- Servers which require more than 4 CPUs
- Servers with external devices attached
- Heavily loaded Citrix servers.
His main points for others considering similar projects were that:
- Providing the infrastructure is in place, migration is straightforward (BT are currently hitting 50-60 migrations per week) with the main activities involving auditing, workload management, downtime and managing customer expectations.
- The virtual infrastructure is truly providing capacity on demand with the ability to deploy new virtual machines in 11-15 minutes.
In another presentation, Peter West, one of Nationwide Building Society’s Enterprise Architects, outlined Nationwide’s server virtualisation strategy. Like many organisations, Nationwide is suffering from physical server sprawl and increased heat per unit of rackspace. As a major user of Microsoft software, Nationwide had previously begun to use Microsoft Virtual Server; however they moved to VMware ESX Server in order to benefit from the product’s robustness, scalability and manageability – and reduced total cost of ownership (TCO) by 35% in doing so (Virtual Server was cheaper to buy – it’s now free – but it cost more to implement and manage).
Nationwide’s approach to virtualisation is phased; however by 2010 they plan to have virtualised 85-90% of the Intel server estate (production, quality assurance/test, and disaster recovery). Currently, they have 3 farms of 10 servers, connected to EMC Clariion storage and are achieving 17-18:1 server consolidation ratios on 4-way servers with data replication between sites.
Peter West explained that Nationwide’s server consolidation approach is more than just technology – it involves automation, configuration and asset management, capacity on demand and service level management – and a scheme known internally as Automated Lights-out Virtualisation Environment (ALiVE) is being implemented, structured around an number of layers:
- Policy-based automation
- Security services
- Resource management services
- Infrastructure management services
- Virtualisation services
- IP networks
With ALiVE, Nationwide plans to take 700 development virtual servers, 70 physical development servers, a number of virtual machines on a VMware GSX Server platform and 500 physical machines to VMware Infrastructure 3, addressing issues regarding a lack of standard builds, backup/recovery, limited support, and a lack of SLAs along with a growing demand from development projects, to allow self service provisioning of virtual machines via a portal.
At the risk of sounding like an extension of the VMware marketing department, hopefully, these two examples of real-life virtualisation projects have helped to illustrate some of the advantages of the technology, as well as some of the issues that need to be overcome in server virtualisation projects.