2 QLogic vs. Emulex HBA Performance Benchmarking with Oracle Workloads
WP06005 SN0032007-00
A
Introduction
The objective of this exercise was to test and compare the I/O performance of QLogic and
Emulex 4Gb Fibre Channel HBAs in a representative selection of database-driven
workloads, a configuration arrived at through a combination of customer requests as well as
common reference within Oracle user groups.
•
An
OLTP workload
with a read:write ratio of 70:30
•
OLTP applications are high throughput, insert/update-intensive and typical
for companies having large transaction processing applications, such as
airline ticketing, large order-entry, and banking applications.
•
A
Web workload
with read:write ratio of 90:10
•
Most database-backed Web applications have high throughput
requirements and are read-intensive, but not exclusively so. Web
workloads mix a variety of content and data-rich applications ranging from
e-commerce and banking applications to CRM and self-service intranet
applications.
•
An
OLAP workload
with a read:write ratio of 100:0
•
The typical applications of OLAP are in business reporting for sales,
marketing, business performance management (BPM), budgeting and
forecasting, financial reporting and planning.
For our tests, the key measures of HBA I/O performance, especially when used in a
database application, were one of the following:
•
The ability to handle multiple data requests, and the volume of data or throughput
as measured in megabytes per second (MBps) with an increase in the number of
outstanding IOs.
•
The ability of the HBA to scale across multiple ports.
In order to make the benchmarks relevant to today’s businesses, we selected the market-
leading
Oracle® RDBMS
running on Microsoft® Windows 2003™ Enterprise Edition with
Service Pack 1. This is a very popular, highly scalable enterprise configuration in wide
production use for all three workloads mentioned above. Selecting Oracle also allowed us
to use ORION, a tool from Oracle used for predicting the performance of the Oracle
database under load. ORION does this by simulating database I/O workloads using the
same I/O software stack as the RDBMS. This tool gives users the capability to decide the
kind and amount of load, varying from small sequential and random with user-defined data
blocks.
Testing performance
- To test the real-world performance of the HBAs, indicative of use
within small or medium scale enterprises, we used a market-leading
EMC Clariion
CX300™
storage array.
Testing scalability
- To test the maximum throughput of the HBAs, we also conducted the
test using a
Texas Memory Systems RamSan-400™
solid-state storage subsystem.
Solid-state storage dramatically reduces the latency that comes from using disk-based
storage, allowing us to measure the true horse-power of the HBA.
