Disk tracing revisited
Q: In your October 1996 column and the second edition of your book there is a guide to using TNF probes to trace disk accesses. But when I try to use it there are too many I/Os in the trace, and it gets complex. Are there any tools that can help?
A: There are two tools that can be used with TNF trace data. The generic tool is called tnfview, and is available from the http://opcom.sun.ca Web site. I also persuaded Richard McDougall to write a simple disk trace-specific tool that has both command-line and graphical versions. In this column, I'll show you how the tool works and look at a few sample disk access patterns.
Trace disk tool
This tool takes the TNF disk trace records and matches them up in pairs for the start and end of a disk transaction. It can see the time a request occurred and the block that was requested, so a direct trace of seek distance is available. The tool was built in a day or so and hasn't been developed since, so it's a bit crude and is missing a few features -- but it's far better than matching up TNF traces by hand! Richard McDougall wrote the code, and the initials, TD, reminded him of the Tasmanian Devil, so the tool is now called the taztool and the package you install is called RMCtaz.
The kernel has many TNF trace points; they can only be enabled if you have root permissions, so these tools must be run as root. These are "safe" tools to run as they are completely read-only, user mode tools that read from supported interfaces.
The two probe points used are sdstrategy, which is the entry point for scheduling a new disk read or write, and biodone, which signals completion. Allocation and lock down of memory to hold the I/O data is done between these two points, and the virtual memory system can be a source of delay, so the time you get is a little longer than the actual disk subsystem response time. The VM system delay is worse when the system is very short of memory and the request has to wait for the page scanner to reclaim memory before it can allocate a read buffer. Normally, the VM system delay is much shorter than the disk I/O delay itself, so it can be ignored.
The RMCtaz tools pick out activity to a single disk device at a time, discarding all the other TNF records. If you want to process more complex data for multiple devices, the tnfview tool can be used instead.
The RMCtaz package installs three files in /opt/RMCtaz. One is a README file, the other two are the text-based taz command and the Motif GUI-based taztool command.
Running the text-based taz command
This command needs to be told which disk device to access and monitors only one disk partition at a time. If you forget to become root before you run
Symantec Backup Exec 12 and Backup Exec System Recovery 8 deliver industry leading Windows data protection and system recovery. Download this whitepaper to find out the top reasons to upgrade and how to get continuous data protection and complete system recovery.
Data and system loss — from a hard drive failure, malicious attack, natural disaster, or simple human error — can happen anytime. Don’t leave your business vulnerable. Make sure you have a secure recovery strategy in place. Symantec's latest backup and system recovery technology can efficiently restore critical applications, individual emails and documents and even restore your entire system in minutes in the event of a loss.
Businesses face a growing challenge to ensure that the IT environment is properly protected. Backup Exec 12 integrates with other applications in the Symantec family of products, to complement your current data protection strategy, keep your data securely backed up and make it recoverable when you need it most.
