Factors affecting direct attached storage device performance in the application layer

4
The realized performance of a disk drive depends heavily on the nature of the workload. Performance
varies when accessing large blocks of sequential data or small blocks of random data. All rotating media
disk drives have varying levels of I/O request re-ordering, called optimization. This optimization reduces
the combined seek and rotation distance of several outstanding I/O requests that the drive has in its request
queue. The more requests in the drive’s queue, the better the ability to optimize.
From a system perspective, the seek distance to random data can also be minimized by maintaining
contiguous files on the disk by using appropriate system utilities. Disk file fragmentation can significantly
degrade both random and sequential performance.
Solid-state I/O performance
SSDs and HDDs use SAS or SATA protocols to interface with the host system, but SSDs store and retrieve
data in flash memory arrays rather than on spinning media. SSDs have no seek or rotational latency time.
They address any sector of the NAND flash directly in 0.1millisecond. SSD latency includes the time for
memory access and transfer combined with controller overhead.
SSDs excel at random read operations, where their performance can be more than 100 times better than
that of spinning media drives. SSDs perform random writes at least 25 times faster than a comparable
15K rpm HDD. This means SSDs provide improved application performance.
Table 2 compares how a single HDD and SSD perform as a function of I/O operations per second (IOPS)
as part of internal HP testing.
Table 2: Performance measured in IOPS for typical single HDDs and SSDs
All
numbers
±5%
Enterprise Drives
(HDD
15K rpm)
Midline Drives
(HDD
7.2K rpm)
Enterprise SATA
Value/Boot
(SSD)
Enterprise SATA
Mainstream (SSD)
Enterprise SAS
Performance
(SSD)
Capacity
300 – 900 GB
500 – 3000 GB
100 GB
400 GB/200 GB
200 GB
IOPS –
100%
random
writes
340
140
7,000
10,000
15,000
IOPS – 100
% random
reads
380
130
30,000
32,000
43,000
IOPS
random
read/write
(70%/30%)
370
137
17,000
19,000
26,000
Wear protection technology
NAND Flash devices use semiconductor technology that supports a finite number of data writes to the
device, defined as the Maximum Lifetime. Wear-leveling algorithms maximize the endurance or life span of
SSDs. This technology re-maps logical memory blocks receiving frequent writes to different physical pages.
It evenly distributes erasures and rewrites across the storage medium. A pointer array on the SSD controller
contains the logical-to-physical map.
HP ProLiant Gen8 servers come with HP SMARTSSD Wear Gauge to maximize SSD media utilization and
eliminate unplanned downtime. To view the HP SMARTSSD Wear Gauge, run the HP Array Configuration
Utility (ACU).
The HP Wear Gauge utility monitors and reports the percentage of a drive’s life cycle used and the amount
of life remaining under the workload-to-date. The utility notifies you through OS event logs, SNMP Storage