Intel® Performance Evaluation and Analysis Kit (IPEAK)
Storage ToolKit FAQs
General
Analyze Disk
AnalyzeTrace
Charts that show disk location don't show the boundary between partitions
With read caching off, the Read CPU Utilization numbers do not show up after the first bar graph (1.5% @ 100 I/Os per second)
Receiving unrealistic transfer rates on different types of drivers
Win32 Tracing Kit
General
I know that the Intel Performance Evaluation and Analysis Kit (IPEAK) Storage Toolkit supports Windows NT* File System (NTFS) and File Access Table (FAT). Does it support Universal Disk Format (UDF)?
The IPEAK Storage ToolKit tools are file-system independent. The AnalyzeDisk tool accesses the disk through the raw disk interface and entirely bypasses the file system. Any storage device that one can assign a drive letter to under Windows NT should work. The Win32 Tracing Kit will show the differences in the file systems as it captures disk access data on a logical block level after it has been translated by the applicable file system.
The Microsoft Excel macro fails with an obscure error message while formatting a graph during the loading of an .SPK file.
Some graphing formatting commands fail in the Excel macro if the system on which the Excel macro is executing has no printer configured; this is not a problem with IPEAK. This is consistent with both Excel 95 and Excel 97. The system on which the Excel macro is executed must have a printer configured. If the system has no printer attached, a dummy printer should be configured in the control panel. The only requirement is for Excel to believe there is a printer configured.
How do I load the tracing driver earlier/lower in the storage stack? (Currently, it loads at load level 4.)
The driver can be loaded earlier/lower in the storage stack by changing its TAG value in the registry from 4 to 2. The registry key can be found in the SERVICES region of the CurrentControlSet portion of the registry, and resides under the heading NTWIOT.
AnalyzeDisk
When I run AnalyzeDisk I get the message: "This program requires Windows NT 4.0 build 1381 service pack 2 or later. Version = 4.0 Build = 1381 Service Pack = 0".
This is an operating system issue and not a Storage Toolkit problem. Please contact Microsoft for a copy of their latest service pack for Windows NT 4.0.
AnalyzeTrace
Charts in AnalyzeTrace don't show the disk to which they relates
The disk selected for all the analyses is the one selected in the drop-down box on the "Raw Trace File" tab. No information about the disk other than its Operating System (OS)-assigned handle (shown in the drop-down box as its unit number) is known or contained in the trace file.
Charts in AnalyzeTrace don't show whether it relates to reads, writes, or both read and writes
Each trace analysis chart relates to the trace as filtered in the "Raw Trace File" tab. Whether the analyses relate to reads or writes is based on how the user filters the trace using the options in that tab.
Those charts that contain what is really an "and above" display at their right-hand end, should give the user an option of either scrolling horizontally or of compressing the scale to be able to show all data. Examples include the service times, queue length, and the transfer sizes charts. The transfer sizes chart stops at 64 sectors. Transfers are often greater than 32KB.
All the charts already have the option of allowing the user to change the scale of the X-axis. Double-click the chart to bring up the properties dialog box, and change the X-axis settings as desired.
Charts that show disk location don't show the boundary between partitions
Trace files contain only the sequence of disk-logical block numbers and their O/S unit numbers, and there is no disk geometry information in the trace. It is therefore not feasible to include the information on the partition boundaries suggested, since this information is not contained in the trace file.
With my read caching off (giving my lowest processor utilization), my Read CPU Utilization numbers do not show up after the first bar graph (1.5% @ 100 I/Os per second). After saving the data to a file, I have looked in the .SPK file, but the numbers do not show up there either. Is there a minimum numerical value for this plot?
When testing your device with the read cache disabled, your device is not able to sustain an I/O rate higher than 100 I/Os per second. In order to sustain a reasonably high I/O rate without slipping revolutions, the read cache may not be disabled, since doing so cripples the performance of the drive to the point where it is not able to sustain a higher I/O rate than the 100 limit. In short, in your configuration, your drive, is performing so poorly that it is unable to perform more than 100 I/Os per second, which is why the test is not able to measure the utilization at higher I/O rates.
I have noticed that for "CPU Read Utilization" I get unrealistic transfer rates on different types of drives. On the "Read Cache Miss Service times" I see only a 2 ms difference between the 0 and 64 sector measurements.
The CPU utilization test measures the relationship between processor utilization and I/O rate. In order to maximize the "dynamic range" of the I/O rate sweep, we use an access sequence that the device can support high I/O rates for. To that end, we use sequential accesses and sweep the inter-arrival times. For sequential I/Os most devices will support very high I/O rates. A 2ms difference over 64 sectors for cache miss computes out to a transfer rate of about 16MB/s, which does not sound right for most commodity desktop drives. In PIO mode 4 this would be pretty close to the cache HIT transfer rate. We have seen anomalies in the transfer rate relationship before, where the service time does not have a very linear relationship with transfer size producing an effect similar to what you report.
Win32 Tracing Kit
Capturing traces with AMI’s MegaRAid* 428 board.
Depending on the configuration of the RAID controller, the captured trace may not appear to have any references to the individual drives, but rather make the complete array appear as a single drive. Since the tracer captured physical disk accesses (or what it thinks are physical disk accesses), it will make no distinction between various drives that make up the RAID array nor will it make distinctions between the various logical drives that may reside on a large array. If the RAID controller is functioning like most that we are aware of, it will present the array as one large physical drive, which is how the tracer will see it, regardless of the number of logical drives that are created on it. One other tip, the tracing device driver currently loads at load level 4 (after ftdisk and most other drivers are loaded). The driver can be loaded earlier/lower in the storage stack by changing its TAG value in the registry from 4 to 2. The registry key can be found in the SERVICES region of the CurrentControlSet portion of the registry and resides under the heading NTWIOT.
After installing the IPEAK Storage Toolkit, in trying to run the Win32 Tracing Kit and it came up with the error "The companion device driver could not be installed. Refer to ‘Install Driver’ under ‘Advanced Settings’ in online help for assistance." How do I resolve this?
In order to install device drivers on Windows NT, you have to be logged in with an user ID that has administrative privileges. To resolve this problem, log in to the computer that you wish to install the IPEAK Storage Toolkit on with an administrative user account.
