IBM TotalStorage DS6000 Series: Performance Monitoring and Tuning

Chapter 7. Open systems servers - UNIX 227

7.7 Testing and verifying DS6000 storage

To characterize storage performance from a host perspective, we enter into a

multidimensional discussion involving considerations that include: throughput (IOPS and

MB/s); read or write; random or sequential; block size; and response time. Here are a couple

of rules that will help characterize these terms:

򐂰 Throughput or bandwidth is measured in two separate and opposing metrics: Input/Output

operations per second (IOPS) and data transfer rate (MB/s). Generally, random workloads

are characterized by IOPS and sequential workloads by MB/second.

򐂰 Maximum IOPS are experienced when moving data that has a smaller block size (4 KB).

򐂰 Maximum data transfer rates are experienced when moving data with larger block size

(1024 KB).

򐂰 Writes are faster than reads because of the affects of the disk subsystem’s cache.

򐂰 Random workloads are a mixture of reads and writes to storage.

򐂰 Sequential workloads generally have higher MB/s throughput than random workloads.

See Chapter 12, “Understanding your workload” on page 407 for an understanding of

workloads.

The UNIX dd command is a great tool to drive sequential read workloads or sequential write

workloads against the DS6000. It will be rare when you can actually drive the DS6000 at the

maximum data rates that you see in published performance benchmarks. But, once you

understand how your total configuration (for instance, a DS6000 attached with 4 SDD paths

through two SANs to your host with 2 HBAs, 4 CPUs, and 1 GB of memory) performs against

certain dd commands, you will have a baseline from which you can compare things like

operating system kernel parameter changes or different logical volume striping techniques in

order to improve performance.

In this section, we will discuss how to:

򐂰 Determine the sequential read speed that an individual vpath (LUN) can provide in your

environment.

򐂰 Measure sequential read and write speeds for file systems.

While running the dd command in one host session, we recommend you use the UNIX

commands and shell scripts presented earlier in this chapter. We will assume that, at a

minimum, you will have the AIX nmon tool running with the c, a, e, and d features turned on.

Below, we will be running lots of different kinds of dd commands. If, at any time, you want to

make sure there are no dd processes running on your system, execute the following

kill-grep-awk command:

kill -kill ‘ps -ef | grep dd | awk ‘{ print $2 }’‘

7.7.1 Using the dd command to test sequential Rank reads and writes

To test the sequential read speed of a Rank, you can run the command:

time dd if=/dev/rvpath0 of=/dev/null bs=128k count=781

Caution: Use extreme caution when using the dd command to perform a sequential write

operation. Ensure the dd is not writing to a device file that is part of the UNIX operating

system.