Network Computing is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Server Memory vs. SSD

When I write or talk with someone about solid-state disks (SSDs), a common question is why shouldn't I buy more server memory instead? There are several reasons why an SSD is a better investment than putting more memory in your servers.

There are the specific implementation differences. SSD solutions are non-volatile. Flash-based SSD is persistent by default;, DRAM-based SSD usually has either a battery backup with hard-drive copy-out or newer solutions have flash memory as the backup. The use of flash- or DRAM-based SSD allows for quick repopulation of the SSD if there is a failure. Also, a SSD is isolated from the server. Your server loses its memory if there is a loss of access to external power. If your application server crashes, all the data in memory crashes with it.

From a utilization perspective, the common use of memory in a server is to increase the size of the cache, and this also applies to memory added to a storage controller, by the way. In either case, you are limited to the accuracy of the cache and, depending on the workload, will be at the mercy of a cache miss. Once a miss occurs, you are waiting for the slow mechanical drive mechanisms to respond. Large cache is rendered useless by large data sets that require very random access.

While accessing data from a flash SSD is not as fast as accessing it from server cache, flash SSDs do offer the performance advantage of never generating a miss that requires access to the mechanical drive, essentially allowing you to load your entire data set onto the flash SSD. If your performance requirements demand cache memory-like performance all the time, then DRAM-based SSDs can accomplish that task, outperforming cache memory.

There are also cost considerations. Server memory is seldom transferable to the latest generation of server; each new server requires faster and faster memory. Also, the more memory you put in the server the denser the chip that has to be purchased because there is a limited amount of space. SSDs can use the most economical density available and they can be carried forward with the new server investment.

  • 1