|
1 Introduction to blade computing
1.1 Data centre background and the problems today
Traditionally, IT has developed along the lines of a number of silos with a wide range of legacy systems that have built up over time. Applications will frequently have been built on different hardware, which in turn will run different operating systems and access a wide range of data sources. A typical IT installation will have a diverse mixture of hardware, many of which reproduce the same functionality of hardware elsewhere in the infrastructure. Quocirca research has shown that CPU utilisation may well be under 10% on average, with storage utilisation at under 30%. Such redundancy of resources leads to consequent high overheads in power use, effective cooling and maintenance costs. With such infrastructure "stacks" being monolithic, it is very difficult to make better usage of these resources - there is simply not enough time for re-provisioning the resources with new operating systems, application servers and the application itself to make better usage of the resources.
Servers are also saddled with a historical architecture - standard motherboards with standard layouts. Direct Attached Storage (DAS), which can not easily be shared, is typically used for at least part of the organisation's storage, and this means single points of failure and high heat sources. Software is installed using optical devices such as CD-ROMs or DVD drives, with only intermittent usage still requiring power availability to be built into the server power supply. Resilience is built in by having multiple pieces of each hardware resource. Each part of the hardware requires of its own power capability, which builds up to the need for dedicated high-output power supplies running well below their capabilities, with internal cooling capabilities designed to meet the peak heat dissipation of the overall unit, combined with large areas of empty space to ensure that the servers do not overheat.
Previous attempts to minimise the size problems of servers have concentrated on trying to condense what there is into smaller boxes - the motherboards, the storage, the optical drives and so on remain, but we try to fit them into smaller units that can be stacked into a standard 19 inch rack, similar to those used within the old telecoms wiring closets. Although this has brought some benefits, power utilisation and heat dissipation remain major problems.
In many data centres, the main constraint to growth is no longer the actual real estate required for the servers - compute density has increased such that most data centres are not fully populated. The main constraints are the provision of uninterruptible power and the removal of the heat being generated within such small, but highly powerful, computers.
Therefore, we now have large data centres, where the pressures being placed upon the IT department are such that management of the existing resources and any sort of growth are both difficult. The plethora of hardware combined with the heat and power constraints brings many problems:
- hardware, particularly server and storage, use is inefficient and costly - more kit is bought in than necessary; - new applications are more costly to develop than they should be - the simplest and quickest option is often to put them on new servers rather than distribute them around, adding to cost in the long term; - the overheads on the IT department are higher than they should be - more staff are needed to maintain and troubleshoot the excess kit; - overall systems management is difficult to impossible, yet a fault in one application silo can have a domino effect on other servers in the data centre, due to power and heat constraints.
1.2 What is blade computing and how can it help?
In blade computing, a single chassis contains a number of ?blades'. Each blade may contain one or more processors and memory, and may also contain its own storage capabilities, along with network connections. However, each function may be kept separate - CPU blades may be installed as a collection of resource, with storage blades being elsewhere in the chassis, or indeed being kept completely separate as Network Attached Storage (NAS) or as a Storage Area Network (SAN). A blade chassis is intelligent, enabling the blades to share data and power, and the whole unit can be engineered for optimum cooling. The chassis allows resilience to be built in using multiple power supplies, redundant networking blades, automatic failover from one blade to another on failure and so on. The chassis manages much of the necessary wiring between blades via high-speed data busses, so lowering the amount of external wiring required.
|
