Electrical computers and digital processing systems: virtual mac – Task management or control – Process scheduling
Reexamination Certificate
2000-09-29
2004-11-30
An, Meng-Al T. (Department: 2127)
Electrical computers and digital processing systems: virtual mac
Task management or control
Process scheduling
C718S100000, C718S102000, C718S106000, C710S200000
Reexamination Certificate
active
06826754
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Technical Field
This invention relates to a method for eliminating or reducing hangs during a shutdown or a panic of computer systems and improves the validity of debugging data acquired from the panic shutdown.
2. Description of the Prior Art
Arguably the most important part of a computer system is the data stored on it. This data is almost exclusively stored on offline storage devices such as disks. It seems to be a continuing trend that the speed of computers remains ahead of the speed of offline storage, and for these reasons most implementations of computer systems use a caching mechanism to keep frequently/recently accessed data in memory rather than on disk. There is a risk with this caching, namely that if anything prevents the cached data from being written to offline storage, the data will be lost. As an example, a sudden loss of power can result in cached data being lost. Another source of cache loss is a failure of the operating system (OS) that results in sudden shutdown without flushing (writing to offline storage) of the caches.
Symmetric Multi-Processor (SMP) computer systems use mutual exclusion (mutex) mechanisms to coordinate (prevent collision) of the activities being performed by the various processors (threads of execution). The integrity of these mutexes must be maintained, or severe problems will result including data corruption. Additionally, to prevent problems such as deadlock (system hangs) or system slowdown, it must be ensured that threads which acquire mutexes, do so for a minimal time and according to well-defined rules set forth by the designers of the code. During normal system operation, once the code design has been completed and the implementation sufficiently debugged, these mutexes tend to operate quite well. However, there are some cases where the operation of these mutexes can be disrupted, most notably during system shutdown but also as a result of “bugs” (in both hardware and software), and in general instances of system failure. Typically, the most reliable mechanism for recovering from these disruptions, is to shutdown the system and restart it from scratch.
Panic is a special case of OS shutdown. During a normal, orderly shutdown, the OS has the luxury of gracefully terminating applications and subsystems, including database systems and file systems, which results in complete data preservation (i.e. zero data loss or corruption). By definition, a panic is the expedited shutdown of the OS and system after detecting a failure of hardware or software. This creates a hostile environment in which the panic shutdown must operate. There are two main goals of a panic:
1. Minimize corruption and data loss, mainly by synchronization of in-memory (cached) data with off-line storage (disks).
2. Accurately preserve the state of the software and hardware in order to facilitate diagnosis of the root cause of the problem.
Note that these two goals may be in conflict under certain circumstances, for example, if a panic is the result of a failure in the disk subsystem. In fact, some circumstances may make it impossible to accomplish either, or both, of these goals.
Another intent of a panic is to restart the system with the hope that a fresh start will allow the system to run more successfully. To this end, the time taken by the panic shutdown, core dump, and reboot is critical to the availability of the system and must be minimized. With the unpredictable nature of the system during a panic, there are many impediments to completing the shutdown that may be encountered. One significant impediment is due to the fact that applications have not terminated (gracefully) and thus have not released the mutexes that they may have been holding. The panic shutdown must resolve these mutexes without adding to data corruption, more specifically it must ensure that “dead” mutexes are broken promptly while mutexes held by legitimately running threads are not “stolen”.
Aahlad, U.S. Pat. No. 5,907,675, provides a method for managing deactivation and shutdown of a server. The patent teaches a method for a server to gently exit without actually terminating the server's clients. There is a lock-up flag that is used to prevent any new clients from joining. The invention implements an orderly and predictable server deactivation and/or shutdown. The invention utilizes an “usher” that continuously maintains a transaction counter indicative of the number of clients that are actively utilizing services.
Hapner, U.S. Pat. No. 5,940,827, provides a method for using mutexes to protect the shutdown time commit phase of a database and the shared database cache. The system informs the shutdown thread that the time commit thread is going away. The mutex is created to correspond to a piece of code.
However, while Aahlad and Hapner disclose shutting down an application, the present invention relates to shutting down the operating system that supports these applications. The difference between the prior art and the present invention is analogous to turning off one's television set verses shutting off the power for the whole house.
Therefore, there remains an unmet need for expediting panic shutdowns by eliminating or reducing hangs while minimizing data corruption/loss and processing system state information.
SUMMARY OF THE INVENTION
One aspect of this invention is an improved mechanism for dealing with mutual exclusion (mutex) structures in a computer system which can have at least one possible state transition. The mechanism includes an identifier, an indicator, and a mutex handler. The identifier identifies the owner of the mutex, which is preferably a processor or process. The indicator shows whether the mutex was acquired before or after the state transition. Where the state transition is from normal operation of the computer system to a “panic” shutdown, the indicator shows whether the mutex was acquired pre- or post- panic. The mutex handler is responsive to the identifier and the indicator, and in the preferred embodiment includes first routines for use before the state transition, and second routines for use after the state transition.
Another aspect of the invention is a method for handling a mutex after a state transition in the computer system. The method determines whether the mutex was acquired before or after the state transition, and handles the mutex differently depending upon whether the mutex was acquired before or after the state transition. In a preferred embodiment, the mutex determines from a data structure whether the mutex was acquired before or after the state transition.
Yet another aspect of the invention is an article of manufacture comprising a computer-readable signal bearing medium such as a recordable data storage medium or a modulated carrier signal. Means in the medium determine whether the mutex was acquired before or after a straight transition in the computer system, and handle the mutex differently depending upon whether the mutex was acquired before or after the state transition.
The invention thus adapts to panic shutdowns and other state transitions in a computer system, distinguishing between pre- and post-transition mutexes, and handling them differently so as to minimize corruption and data loss and accurately preserve a system state. Other features and advantages of this invention will become apparent from the following detailed description of the presently preferred embodiment, taken in conjunction with the accompanying drawings.
REFERENCES:
patent: 5765151 (1998-06-01), Senator
patent: 5822512 (1998-10-01), Goodrum et al.
patent: 5864653 (1999-01-01), Tavallaei et al.
patent: 5907675 (1999-05-01), Aahlad
patent: 5940827 (1999-08-01), Hapner et al.
patent: 6006247 (1999-12-01), Browning et al.
patent: 6021262 (2000-02-01), Cote et al.
patent: 6223204 (2001-04-01), Tucker
patent: 6243788 (2001-06-01), Franke et al.
patent: 6378124 (2002-04-01), Bates et al.
patent: 6546443 (2003-04-01), Kakivaya et al.
patent: 6636950 (2003-10-01), Mithal et al.
patent: 6662364 (2003-12-01), Burrows et al.
IBM T
Ali Syed
An Meng-Al T.
International Business Machines - Corporation
Law Offices of Michael Dryja
Raissinia Abdy
LandOfFree
Method for eliminating or reducing hang conditions in... 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 for eliminating or reducing hang conditions in..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for eliminating or reducing hang conditions in... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3314022