LOSTCIRCUITS

A different strategy for writes is to write all data to a dedicated area within the system memory and then schedule them for the most economic write pattern to the drive. It is a very straightforward strategy utilizing part of the system memory as dedicated write cache. An additional advantage is that any further modification of the data will not even have to access the drive itself but can be executed in memory within the resident copy of the data and the drive will be updated whenever it is necessary. In other words, what we are looking at is the equivalent of a RAM-disc or the reversed principle of the "Virtual Memory" strategy where the HDD needs to handle the data overflow in case there is not enough system memory available.

In that case, the "virtual drive's" data bandwidth will be similar to the memory bandwidth, which is an order of magnitude higher than what even the fastest drive can accomplish. The data on the HDD media are updated at the rate supported by the drive itself, however, data access itself, since it occurs from memory will bypass the drive itself completely and, therefore, be ultra fast. Issues associated with this strategy are first, the same problems as with caching data in non-volatile memory, namely the risk of data loss and, second, the fact that only recurrent data can be cached. On the other hand, using system memory as write buffer allows more efficient use of data, in other words, the data transfer to the drive itself is not influenced to the same degree by interrupts and PCI latencies or bus bottlenecks as in the case of a standard configuration. Overall, it is a nifty concept but, as outlined above, not without risks. The question is, how much of a difference does it actually make in terms of real world performance.