h

IO

mon

™

  The I/O Performance Monitor

Home  |  Products & Services  |  Contact  |  About hyperI/O^sm

Backgrounder

Rants from hyperI/O

DIOR Big Picture

IOPM Big Picture

The arena of storage I/O operation performance is one fraught with varied thoughts, opinions, and claims.  Our own bias is one with a primary focus upon empirical metrics, which reflect actual observations of storage I/O operations that have been captured upon your very own systems during the normal, everyday usage of your particular applications.

The commentary below briefly discusses several key topics related to the analysis and understanding of storage I/O operation performance.  Also see the "Key Storage Performance Questions".

Product Info

Features

Architecture

Screen Shots

Download

Support

How Fast is Fast Enough?

Certainly disk storage I/O performance has long been a significant bottleneck within computer systems and their overall performance.

Fortunately, the recent more-widespread availability of Solid State Disk (SSD) technology has helped to address this issue in large measure.  For example and in terms of raw performance potential, many SSDs can support I/O operations-per-second (IOPS) rates that are several orders of magnitude greater than mechanical hard drives.

However, greater performance usually demands a premium.  Simply stated, added performance in a product typically costs more.

So where cost is a concern (as is the case for most folks), basic questions arise as to how much of this "raw performance potential" do you actually need and will you actually use.  Why pay more for what you can't or won't use?

Capturing the actual "I/O profile" (based upon empirical metrics that reflect normal, everyday usage) of your particular computer system is a key starting point for evaluating your actual disk I/O performance usage and needs.

And the ability to quickly and easily collect, display, and export such empirical metrics (including upon the simple "speeds and feeds" basis along with the "DataTransferred/Time Index" metric for comparing relative I/O performance) is one of the hallmark features of hIOmon.

Benchmarks:  The Good and the Not-So-Good

When used within a well-controlled setting and manner, storage I/O performance benchmarking tools can provide valuable insights into maximum innate performance capabilities, comparative disk advantages/shortcomings, etc. 

But then there is that "great leap across the abyss" when trying to correlate benchmarks and their particular results to the normal, everyday usage of your specific computer systems and applications.

For example, some benchmarks attempt to portray the typical I/O activity of typical applications.  But are the actual I/O profiles of your particular applications ”typical”?

The empirical I/O operation performance metrics made available by hIOmon can help you assess how “typical” your applications are and how representative a particular benchmark might be of your specific applications and their I/O usage.

hIOmon can also be used to determine the appropriate input parameters (e.g., read/write ratio, queue depth, random/sequential access ratio, etc.) for use with benchmarking tools that support configurable options related to the type of I/O operation activity that is generated by these tools.

And beyond the usual synthetic benchmarking tool approach, you can use hIOmon to perform a "before" and "after" comparison based upon empirical I/O operation performance metrics when evaluating proposed improvements, such as the use of SSD, faster hard drives, iSCSI, additional cache or system memory, faster/more processors, file defragmentation approaches, virtualized storage, virtual machine consolidation, and more.

Observations at the Other End of the Cable

Discussions about disk I/O performance (particularly storage vendors claims) are often centered around the "speeds and feeds" of the disk devices themselves.

But ultimately it is the performance at the other end of the cable (i.e., at the host -- and particularly at the applications themselves) that is of uppermost interest to users.

That is, most folks buy fast (or faster) disk drives not simply to possess fast drives, but rather to gain a beneficial, cost-effective, and notable improvement in the overall performance of their computer systems and the associated tasks/applications.

hIOmon provides the ready means to actually observe storage I/O operations and their performance at "the other end of the cable".

By observing the actual file I/O operation activity performed by your particular applications, hIOmon can both measure and monitor disk I/O operations up at the application level upon an individual, specific file basis.

The hIOmon software can also optionally monitor "raw disk" I/O operations, especially at either the “physical volume” and/or the “physical device” levels within the operating system for partitioned/formatted volumes -- and even do so concurrently at all three levels within the operating system I/O stack.

This unique capability of hIOmon helps provide a more complete "top-down" perspective of actual I/O operation performance in comparison to simply a singular focus upon the I/O performance of the disk drives alone. 

hyperI/O^sm -  improving the performance of storage I/O in computer systems