Solid state drive technology for ProLiant servers

5
Wear leveling for increased SSD endurance
Wear leveling is one of the design techniques engineers use to increase the endurance of NAND-
based SSDs. Since NAND-based SLC flash supports only 100,000 lifetime write/erase cycles, the
SSD needs to not erase and rewrite NAND blocks any more than is necessary. But application usage
may require frequent updates or rewrites of some logical SCSI blocks in a SAS/SATA device. Wear
leveling resolves this issue by continuously re-mapping logical SCSI blocks to different physical pages
in the NAND array. Wear leveling ensures that erasures and rewrites remain evenly distributed
across the medium, maximizing the SSD’s endurance. To maximize SSD performance the SSD
controller maintains the logical-to-physical map as a pointer array in high speed DRAM. The metadata
region in the NAND flash array itself also maintains this information algorithmically. These techniques
ensure that the SSD can rebuild the map if you lose power unexpectedly.
Over-provisioning NAND
Design engineers can increase the endurance and performance of an SSD by over-provisioning
NAND capacity on the device. Over-provisioning increases the endurance of an SSD by distributing
the number of writes and erases across a larger population of NAND blocks Over-provisioning also
increases SSD performance by giving the SSD controller additional buffer space for managing page
writes and NAND block erases. On higher-end SSDs, NAND memory may be over-provisioned by as
much as 25 percent above the stated storage capacity.
HP solid state drives for ProLiant servers
HP introduced the first SSDs for servers in 2008. The SSDs were not hot-pluggable and intended for
specific BladeServer environments. In 2009, we introduced the first hot-pluggable SSDs in traditional
drive carriers. These 3 Gb/s SATA SSDs are usable across the ProLiant server line, wherever you
would use a traditional midline SATA disk drive. Unlike PC-based solid state drives, SSDs for servers
meet the higher standards for server storage devices. At the same time, they provide the performance
and reliability characteristics associated with SSDs.
Performance of HP server SSDs
Disk access time, or latency, which is the total time required to retrieve data from the drive, influences
the performance of a traditional disk drive. Disk drive latency is the sum of the seek time, rotational
delay, and transfer time.
With SSDs, there is no seek time or rotational delay. Latency is a function of the memory access and
transfer times combined with controller overhead. Given these facts and the knowledge of how
NAND flash operates, we can make the following suppositions:
 Read operations should be faster on SSDs than write operations, because of the relative slowness of
NAND program (write) operations.
 Random reads on SSDs should be faster than to random reads on disk drives, because the SSDs
has no seek time and rotational delay for each read operation.
Table 3 is a side-by-side comparison of the performance of a 32-GB small form factor HP server SSD
with that of a 15K Midline SAS hard disk drive (HDD). While performance on sequential operations is
comparable between the two drive types, performance on random operations is significantly better for
SSDs. In random read performance, the SSD achieved more than 50 times the performance of the
HDD.