Error detection/correction and fault detection/recovery – Data processing system error or fault handling – Reliability and availability
Reexamination Certificate
2002-11-19
2003-12-30
Baderman, Scott (Department: 2184)
Error detection/correction and fault detection/recovery
Data processing system error or fault handling
Reliability and availability
C714S047300, C709S241000, C717S127000, C717S130000
Reexamination Certificate
active
06671830
ABSTRACT:
TECHNICAL FIELD
This invention relates generally to data processing and, more particularly, to a method and apparatus for analyzing the performance of a data processing system.
COPYRIGHT NOTICE/PERMISSION
A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the U.S. Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever. The following notice applies to the software and data as described below and in the drawings hereto: Copyright© 1997-1999, Microsoft Corporation, All Rights Reserved.
BACKGROUND OF THE INVENTION
In the field of data processing it is a well known problem that software developers usually require a period of time to identify and resolve functional and performance issues in the code they have written or integrated. There can be many reasons for such issues, including the basic system and software architecture; non-optimized and/or flawed coding; the choice of, utilization of, and contention for system resources; timing and synchronization; system loading; and so forth.
Particularly in the area of distributed computer networks, it can be extremely difficult for software developers to observe and isolate undesirable system performance and behavior. A distributed computer network is defined herein to mean, at a minimum, a data processing system that utilizes more than one software application simultaneously or that comprises more than one processor.
For example, a single box or machine which is running two or more processes, such as a data base application and a spreadsheet application simultaneously, fulfills this definition. Also, a single article such as a hand-held computer may comprise more than one microprocessor and thus fulfills the definition.
More commonly, however, distributed computer networks may comprise two or more physical boxes or machines, often hundreds or even millions (in the case of the Internet). A software developer trying to monitor and analyze the operation and behavior of such complex computer networks is faced with a very daunting task.
For example, a developer may be writing or have written a server component that performs credit checks. This software component is used in a larger application that performs order entry processing. There are several other server components in the system (such as inventory verification, order validation, etc.) some of which run on the same server and some which run on a separate server (where the inventory database resides). To complicate matters, each component could reside on a computer system in a different state or country. If the application is not performing or behaving well, the developer needs to figure out if there is a performance or behavioral problem and, if so, be able to determine exactly where the trouble spots are.
In the prior art the developer had to modify his or her application, by writing trace statements in the code and having the application write to a log file what was going on at different places in the network. Then all of the log files would need to be collected, merged, and sorted. The developer would then have to sift through the data in a time-intensive fashion and attempt to determine the performance problem.
There are several serious deficiencies with the prior approach.
One problem is that only instrumented code can be analyzed. That means source code must be modified, recompiled, and re-deployed. This is a serious issue with the widespread use of operating system services and component technology in today's applications. Users are typically unable to recompile operating system and third party components, because they do not have physical or legal access to the source code. When they do have access to the source code, they are still unable to instrument them effectively, because they do not understand the component source code that they do have.
Another problem is that the modifications to code made by developers in an attempt to analyze its performance themselves adversely impact the application's performance. Further, the development of a highly efficient mechanism for recording the application data is non-trivial. Typical implementations involve writing data to disk. Even if the input/output (I/O) is buffered asynchronously, it can have an adverse impact on the application being monitored (e.g. masking actual application I/O).
A further problem is that understanding control flow during transitions is very hard. Typically, in a large distributed application, transitions to separate processes, or to processes running on separate machines, are common, and may happen simultaneously. Since events have to be manually merged by the developer, it is typically hard to determine which suspension in one process corresponds to resumption in another.
An additional problem is that frequently there are a large number of application areas that might need to be analyzed; however, not all of them may need to be analyzed at the same time. Developers who manually instrument their code must incorporate a selection technology to enable different portions to be analyzed. Otherwise, the load of all of the instrumentation has a severe impact on the analysis. This also requires a complex mechanism for developers to specify which information to collect on which machine.
Yet another problem is that for distributed applications, logs from multiple machines (and often multiple logs per machine) must be merged and sorted. Without synchronized clocks, this task is very difficult. As well, if the log files are in different formats (which is likely if they are from different developers or companies), then the data must be translated into common formats.
The result of all the effort described in this section is a very long list of analysis data. Manually analyzing and isolating performance problems from this amount of data is a very complex and difficult task.
One further problem with known performance analysis of data processing systems is that very often such analysis provides opportunities for breaching the data security of such systems.
There exists known performance monitoring software in various forms. Among them is software known as PerfMon software, which is commercially available from Microsoft Corporation. PerfMon software is a utility which, among other things, can provide an indication of the utilization of the computer's central processor unit (CPU) and memory unit. PerfMon software operates by sampling. That is, it tracks continuous data by monitoring a machine and looking at its behavior. It can track the free space on a disk, monitor network usage, and so on, but it cannot gather event-based information, such as what function was most recently started.
There also exist known tools called profilers. These look at a single executing software application and try to understand its performance. They do this either by monitoring the program (in a similar way to PerfMon software), or else they hook into the program they are monitoring and generate “events” each time a program subcomponent (function) commences or completes. Profilers typically have a massive impact on the performance and behavior of an application, because they are intrusive, and they typically require special compiler support. Their data is so detailed that it is normally impractical to use them, particularly in a distributed computing environment such as the one described above.
The Windows NT® PerfMon utility, commercially available from Microsoft Corporation, provides an extensible architecture for the collection and display of arbitrary application and system counters and metrics. Windows NT provides base counters for the system for the purpose of monitoring CPU and memory utilization. It also provides counters for networks, disks, devices, processes, and so forth. Most system objects export counters. Many applications available from Microsoft Corporation (such as
Ferguson William J.
Kaler Christopher G.
Lowell Martyn S.
Sharp Oliver J.
Wahbe Robert S.
Baderman Scott
Microsoft Corporation
Woodcock & Washburn LLP
LandOfFree
Method and apparatus for analyzing performance of data... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for analyzing performance of data..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for analyzing performance of data... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3169242