Understanding endurance and performance characteristics of HP solid state drives

7
 The SSD’s operating temperature when it was in service
 The temperature you store the SSD at after removing it from service.
The data retention period of an SSD is actually greater when you operate the SSD at higher operating
temperatures while it is in service and store it at lower temperatures once you remove it from service. As
an example, an SSD operated at 50°C and stored at 30°C should retain its data for 28 weeks if you
remove it from service at the end of its rated endurance.
The important thing to remember is that an SSD has a limited data retention window once you remove it
from service. This is different from disk drives, which typically retain data for years. If an SSD has used
all of its rated endurance, the only truly safe assumption that you should make when removing it from
service is that it will not retain its data for any significant period.
Understanding SSD performance characteristics
Because Solid State Drives are compatible with the SAS and SATA interfaces, you can measure their
read and write performance using the same tools measuring disk drive performance. But their
underlying storage technology is different from that of disk drives. As a result, their performance
characteristics are also distinctly different. With SSDs, we need to re-examine our assumptions about
storage performance and understand how SSD performance changes in different environments and
under different workloads.
Measuring SSD performance
SSDs are capable of delivering exceptional performance, particularly for random I/Os per second
(IOPS). You can measure SSD performance by using Iometer or other tools to compare it with that of a
disk drive. But you’ll discover that an SSD’s performance can vary significantly each time you run the
same test unless you use the proper methodology. We can attribute these differences to the varying
overhead of the background management tasks associated with the NAND memory architecture.
SSD NAND organization and performance
In addition to fulfilling read/write requests, an SSD controller is executing background tasks to manage
the NAND memory. They include NAND block management to maintain a pool of free blocks, and
data re-mapping tasks associated with wear-leveling. The level of background activity can vary
significantly, due in part to the organization of the NAND data and the type of read/write activity
going on. The changing level of background activity influences SSD performance.
When running benchmarks on SSDs, all of the following are true:
 Performance drops by as much 50% when written data starts to fill a new SSD’s storage capacity. As
the SSD fills with data, the level of background NAND management activity rises dramatically,
increasing overhead and lowering performance.
 Performance drops within the first few minutes of starting a benchmark. The level of background
activity gradually increases because of the read and write activity of the benchmark.
 Performance can increase after running a sequential write test. The test leaves the SSD NAND in a
more organized, less randomized state. This lowers NAND management overhead for the next test.
Pre-conditioning an SSD for accurate performance measurements
To obtain accurate, repeatable performance measurements for SSDs, you should first pre-condition them
to a steady state that reflects how the SSD will operate in most production environments. At HP, we pre-
condition SSDs for benchmark tests using the following procedure: