An approach to storage caching that centralizes the resource and uses Ethernet, IP and Network File System can serve RAM-cached files 10 to 50 times faster than mechanical disk-based approaches.
Delivering high-speed, high-capacity caching as a shared network resource for I/O-intensive requests means data centers can reduce the need to overprovision storage and guarantee QoS for all NFS-based application servers and storage systems.
The need for centralized caching is due, in part, to the fact that it is hard to coherently distribute memory across servers or storage for end-node caching. Typically memory is replicated across devices to balance performance, and each server or storage system maintains a unique memory pool.
But mixed workloads can result in underutilized resources, and servers and storage devices have a finite memory capacity so the need to increase overall memory requires adding systems, and that may result in wasted CPU or storage capacity
An alternative is a centralized storage cache implemented as a network-attached appliance on the IP network that can be used by all application servers. Frequently accessed data can be stored in high-speed memory and delivered to application servers in a fraction of the time compared with slower, mechanical disks.
Similarly, centralized storage caching accelerates any NFS storage systems, allowing the investment to be amortized over the data center infrastructure.
Centralized storage caching builds on proven industry standards such as Ethernet, IP and NFS. Architects can call for caching appliances knowing they will be interoperable with all NFS-based application servers and storage systems.
The progression of new Ethernet capabilities, such as support for 10Gbps links, helps ensure that centralized caching solutions will be able to keep pace with other data center advances.
Unlike the limited capabilities of end-node caching, centralized storage caches scale independently of server and storage systems. This enables architects to add caching resources at any time and scale cache capacity to multiple terabytes. Because this capacity can be large enough to cache entire data sets, architects can expect guaranteed response times from high-speed memory and create robust service levels for enterprise applications.
When you're evaluating storage-caching options, look for the ability to support:
* Coherent cache pool. To store terabytes of data, centralized storage-caching appliances should virtualize cache memory across appliances. This clustered approach creates terabyte-sized coherent cache pools that maintain a single namespace, making it easy for any number of application servers or storage systems to make use of the shared resource.
* Parallel I/O. Taking advantage of terabyte-size cache pools requires massive parallelization, whereby a caching appliance can load balance and distribute requests across the pool to accelerate response times.
* Linear scalability. You want to be able to grow cache capacity while increasing overall I/O operations per second and throughput levels, and reduce data retrieval times.
From: Centralized storage caching, Network World, 06/04/07.
Additional resources
