Intel 240gb User Manual Page 7

  • Download
  • Add to my manuals
  • Print
  • Page
    / 19
  • Table of contents
  • BOOKMARKS
  • Rated. / 5. Based on customer reviews
Page view 6
Intel
SSD DC S3500 Workload Characterization in RAID Configurations
December 2013 White Paper
329903-001US 7
4.2 Queue Depth and Latency
LatencyThe amount of time needed to service one outstanding IO to the drive,
measured in milliseconds (ms) or, with SSDs, microseconds (µs).
The Intel
®
SSD DC S3500 Series supports a maximum queue depth of 32 per drive. In a
RAID array, the queue depth is multiplied by the number of drives in the RAID set.
Example: In a RAID 5 set of 8 drives, the maximum total queue depth would be 256 (8
X 32). As more commands are queued in the SSD, average latency is impacted. Our
internal testing indicates that average latency increases sharply with queue depths
beyond 8. However, these high queue depths can increase IOPS with read intensive
workloads.
Obtaining the best performance for a particular application requires balance. The
challenge is to achieve high speed or IOPS at an acceptable latency level. This white
paper presents lower queue depths of 1, 2, 4 and 8 per drive. The results shown
demonstrate favorable speed and IOPS generation without pushing latency to extreme
levels.
4.3 Why Mixed Workload Is Important
Mixed random workloads are predominant in data center and enterprise applications.
Intel SSDs have been deployed in a variety of these applications ranging from content
delivery and video on demand networks, to Internet datacenter portals and database
management servers. Although these applications see unique IO traffic across the
storage drive, there are commonalities in their usage of random and read/write mixed
workloads.
Table 1 shows an overview of transfer sizes, read/write mixes and randomness in
commonly used workloads in data center applications. These are based on commonly
available industry information and information available through such benchmarks as,
TPC-C, TPC-E, TPC-H, and TPOX, which attempt to mimic these real world applications.
Table 1. Typical Mixed Workloads in Data Center Applications
Application Transfer Size %Random %Read
Web-Servers
4KB/8KB/16KB+
~75%
~95%
Exchange Email
4KB
~95%
~70%
Database OLTP
4KB/8KB
~95%
~70%
Decision Support
16KB+
~95%
~95%
Video On Demand
16KB+
~95%
~95%
Search Engine
4KB/8KB/16KB
~95%
~95%
Cache
16KB+
~95%
~95%
Content Delivery Network
16KB+
~95%
~70%-95%
Based on these usage trends, small transfer sizesranging from 4KB to 16KB and
aboveare common in enterprise and data center applications. Also, much emphasis is
placed on random accesses, and although there are varied levels of read and write
Page view 6
1 2 3 4 5 6 7 8 9 10 11 12 ... 18 19

Comments to this Manuals

No comments