Flash Storage: Ready For Writes

Most data centers have embraced flash-based solid state storage to address storage performance problems, but this adoption has primarily focused on read performance issues. Solid state disks (SSD) and solid state systems (SSS) are now ready for the other half of the I/O problem: writes.

Write avoidance remains a goal for most solid state storage solutions because flash memory's life expectancy is based on the number of writes -- the more a solid state device is written to, the faster it wears out. To read data from a flash device, you must write some data to it at least once. SSDs and SSS have, at a minimum, some technology to spread out writes evenly across the available flash cells. This technique, called wear leveling, is commonplace today.

Most solutions also over-provision flash memory to increase overall lifespan. For example, a drive might add 25% or more flash storage to allow the flash controller to spread writes across more flash cells. The problem with this is that vendors don't give away this unused flash for free, so the more they provision, the more you pay for a flash solution.

[ High-availability appliances combined with software-defined storage could prolong the life of your storage system. Read more at The Impact Of High-Availability Flash Appliances. ]

Wear leveling and over-provisioning are basic table stakes in the flash market today. As I discussed in this recent article, companies have begun to extend these basic capabilities to include the ability to customize the over-provisioning setting. For example, if you are installing an SSD into a very high-write environment, you could sacrifice some useable capacity to get higher life expectancy.

SSD vendors are also adding the ability to adjust the energy charge used to write data to the flash. The softer the charge, the easier it is on the flash cell and the longer it should last. When there is not a lot of write queue depth, they can essentially slow the write down and extend the overall flash life without impacting performance.

As I discussed in this article, vendors can also integrate DRAM, which does not have a write penalty, to help extend the life of flash memory. With this technique, flash memory is front-ended by capacitor-charged DRAM. Writes are sent to the DRAM buffer first, which speeds performance and allows the write to flash to be organized for maximum efficiency, resulting in fewer wasted flash writes.

The combination of advanced over-provisioning and write modification has led to flash-based devices that claim 5 to 20 drive fills per day for five years. For the data center, that is certainly a practical life span.

All this investment in making flash more write-worthy is working: flash is ready for write traffic. As I covered in this recent webinar, many virtual workloads, especially virtual desktops, have a significant write I/O demand. Caching or storing these writes in flash can significantly improve performance as well as virtual machine density.

Now is the time for a shift in how and why SSS is used. Advancements in technology have improved the write-worthiness of the solutions, and they are ready for both sides of the I/O stream.

Our four business scenarios show how to improve disaster recovery, boost disk utilization and speed performance. Also in the new, all-digital Storage Virtualization Gets Real issue of InformationWeek SMB: While Intel remains the biggest manufacturer of chips in the world, the next few years will prove vexing for the company. (Free registration required.)