Intel 10GbE Adapter Performance Evaluation for FCoE and iSCSI

Table Of Contents



Demartek

Intel

10GbE Adapter Performance Evaluation

September 2010

Page 8 of 18

5 – Performance Results

Two sets of tests, Microsoft Exchange JetStress 2007 and Microsoft SQLIO, were run using iSCSI

and FCoE configurations. Both performance and CPU utilization are important, so we tracked

both. The fundamental performance is critical to determine if the storage can support the

application at a particular workload level. The CPU utilization shows how much load that

workload is placing on the CPU.

Performance is a vital measurement when testing converged adapters because some adapters can

sustain higher performance than others for various workloads. The CPU utilization is important to

measure because this shows the load that the adapter driver and interface place on the platform

and will have an effect on the number of simultaneous applications or virtualized operating systems

that can be run on a given platform.

Microsoft Exchange Jetstress 2007 simulates the Exchange Server disk input/output (I/O) load.

This tool, provided by Microsoft, verifies the performance and stability of a storage subsystem and

its suitability for Microsoft Exchange Server. Jetstress is generally used by customers before

deploying Exchange Server to ensure that the storage subsystem can sustain a production workload.

JetStress

The Jetstress configuration we ran represents a company with 5000 employees. These JetStress

configurations used 5000 mailboxes of size 125MB using the “heavy” user profile. There were 8

storage groups spread across 8 LUNs on the storage arrays, all accessed simultaneously. The

JetStress tests were run for a minimum of 2 hours each. Several different thread counts were

specified for each test run, showing increasing workloads.

SQLIO is a tool provided by Microsoft that can be used to determine the I/O capacity of a given

workload. SQLIO can run I/O workloads of various types and is often used by SQL Server

administrators to test a storage subsystem before deploying a production SQL Server application.

SQLIO

We tested 8KB random reads and random writes, which represent OLTP workloads, and we tested

several block sizes of sequential reads and writes that represent workloads such as batch processing,

database backups, database log files, some decision support systems, etc. We varied the queue depth

(number of simultaneous I/O requests) and the thread count (number of “workers” issuing I/O

requests) in order to represent different customer environments and different workloads, from light

to heavy. Though we varied the thread count for these tests, we are reporting the results for one

representative set of thread counts in order to save space in this report. All the SQLIO tests

accessed 8 LUNs simultaneously across the storage arrays.

The following pages provide the performance results for each of the tests. The data shown are the

results of the specific application workloads. Exchange JetStress output provides IOPS and CPU

utilization. SQLIO output provides IOPS and latencies.

Performance Data Comments