Maximizing storage ROI with fast external cache

Computer Technology Review, July, 2004 by Woody Hutsell

Your RAID system has a problem. It isn't that the system controller is going to fail. It isn't that failed disk drives can't be replaced. It's that the system's performance doesn't scale. The RAID, perhaps only a few years old, is throttled by demands on its performance. Yes, additional spindles may add a hundred or so I/Os per second each, but can this keep pace with the increasing demand of applications and users on storage? Most likely, the answer is "no." This means the RAID array is next in line for the dreaded forklift upgrade. For you, ROI is spelled "e-b-a-y".

Fortunately, storage manufacturers are increasingly focused on storage performance. From the disk drive to the distributed RAID, cache (as in RAM) is the secret ingredient in next generation storage performance. But the increase in cache is happening in small quantities in all but the most massive monolithic solutions. What the storage industry has been missing is a way to combine high-speed cache with low-cost RAID storage solutions. Even at the enterprise level, monolithic RAID solutions will become the way of the past. Flexible, high performing, distributed and inexpensive storage solutions offer the best storage return on investment.

The monolithic RAID is the product of a simple progression: small to big. You start with small controllers, small cache and small spindle counts and innovate by creating systems with big controllers, big cache and big spindle counts. This approach has two major disadvantages: expense and performance scalability. For over a decade, if you needed enterprise performance you had no other choices. Increasingly, innovative companies are developing smarter, more scalable solutions. These solutions isolate and distribute the components of their architecture in order to flexibly address application performance requirements.

The first step in the move to distributed storage is separating the controllers from the disk storage. Many RAID storage companies have taken this step. Separating the controllers from the disk storage allows a company to choose the number of ports and the processing power dedicated to their storage and promotes high availability configurations. The second step to distributed storage is separating the cache from the controller. Processing power and cache are the two main tools that RAID companies have for improving storage performance. Most RAID controllers on the market have a limited amount of cache. Therefore, distributing their processing capability may not be enough to offer enterprise performance. In order to meet enterprise performance requirements, companies like Texas Memory Systems offer external storage cache. These external storage cache systems can be flexibly combined with distributed RAID controllers to provide performance where performance is needed.

But how do they work? At its core, an external storage cache is a non-volatile array of fast RAM that interfaces with other storage devices through multiple high-performance interconnects, such as Fibre Channel links. The external storage cache can be flexibly attached to either server(s), switch(es), or storage. A storage device connecting to the external storage cache thinks it is talking to a server. A server talking to the external storage cache thinks it is talking to storage. In other words, the external storage cache is transparent to the storage network. Like the cache in monolithic RAID systems, the cache is battery backed. When external power fails, the unit will run fully functional off of its batteries. After a set amount of time, if power is not yet restored, the system will dump data from the cache to the attached RAID systems. Some external cache solutions will also dump the cache to internal disk drives as an added layer of data protection.

[ILLUSTRATION OMITTED]

Properly designed, an external storage cache will automatically discover storage targets on the network and present these to an administrator as candidates for caching. The administrator then maps the LUNs or storage devices that need caching to the external storage cache. In this way, one external storage cache can be used to accelerate one LUN, multiple LUNs, one RAID, or multiple RAIDs.

External cache systems generally support standard caching modes, such as write-through, write-back, and read-ahead. The most sophisticated of the external cache systems also offer the ability to flexibly configure the cache line size to match the size of incoming data traffic and to control the rate at which data is flushed to attached RAID systems.

External storage cache systems accelerate data access by providing a high-speed data pipeline. By temporarily writing data to fast RAM instead of conventional rotating disks, large cache systems drastically increase burst bandwidth and I/O rates, ensuring that existing storage systems will survive under immediate pressure. Smart caching algorithms "pre-fetch" data that users are likely to access, which increases overall performance by enabling applications to read the most popular data directly from cache.