Electrical computers and digital processing systems: multicomput – Computer-to-computer data routing – Least weight routing
Reexamination Certificate
1995-06-07
2001-06-19
Donaghue, Larry D. (Department: 2154)
Electrical computers and digital processing systems: multicomput
Computer-to-computer data routing
Least weight routing
C709S239000, C709S229000
Reexamination Certificate
active
06249800
ABSTRACT:
CROSS-REFERENCE TO RELATED APPLICATION
This application describes and claims subject matter that is also described in co-pending United States patent application of Jeffrey D. Aman, Curt L. Cotner, Donna N. T. Dillenberger and David B. Emmes entitled “APPARATUS AND ACCOMPANYING METHOD FOR ASSIGNING WORK REQUESTS AMONG A PLURALITY OF SERVERS IN A SYSPLEX ENVIRONMENT”; filed concurrently herewith; assigned Ser. No. 08/476,157, now U.S. Pat. No. 5,603,029 and which is also assigned to the present assignee hereof.
BACKGROUND OF THE DISCLOSURE
1. Field of the Invention
The invention relates to apparatus and accompanying methods for use preferably, though not exclusively, in a multi-system shared data (sysplex) environment, wherein each system provides one or more servers, for dynamically and adaptively assigning and balancing new work and for session requests, among the servers in the sysplex, in view of attendant user-defined business importance of the requests and available sysplex resource capacity so as to meet overall business goals.
2. Description of the Prior Art
Prior to the early-1980s, large scale computing installations often relied on using a single monolithic computer system to handle an entire processing workload. If the system failed, all processing applications in the workload were suspended until the failure was remedied. While a resulting processing delay was tolerated at first, as increasingly critical applications were processed through the system, any such ensuing delays became increasingly intolerable. Furthermore, as processing needs increased, the entire system was eventually replaced with a new one of sufficient capacity. Replacing systems in that manner proved to be extremely expensive and very inefficient. However, at that time, few workable alternatives existed, to using monolithic systems, that appreciably eliminated both these outages and an eventual need to replace the entire system.
To efficiently address this need, over the past several years and continuing to the present, computer manufacturers are providing processing architectures based on a multi-system shared data approach. Through these architectures, multiple large-scale computer systems, each of which is often referred to as a computer processing complex (CPC) or a central electronic complex (CEC), are interconnected, through, for example, a coupling facility or other inter-processor communication mechanism, to permit each such system to gain read-write access to data residing on one or more shared input/output devices, such as a direct access storage device (DASD). The resulting inter-connected computer system is commonly referred to as a “sysplex”. In a sysplex, as with a typical multi-processing environment, a processing workload is generally distributed among all of the inter-connected computer systems such that each computer system is responsible for processing a portion of the entire workload. Conventionally then, each of these systems executes its own portion of the total workload independently of that undertaken by any the other such systems. Owing to the inherent high reliability and highly cost-efficient expansion potential of a sysplex architecture, sysplexes are particularly attractive in handling so-called critical business support applications that involve real-time transaction processing and can tolerate essentially no downtime.
Generally, within a sysplex, separate copies (so-called “instances”) of an application are resident and simultaneously active on more than one of the computer systems, each henceforth referred to as a “machine” to differentiate the physical hardware therefor, and, based upon, e.g., the processing capacity required for the application, often on all or most of these machines.
Certain currently available machines that can be readily incorporated into a sysplex, such as illustratively the Enterprise System/9000 (ES/9000) Series manufactured by the International Business Machines (IBM) Corporation, can each support, if appropriately configured, multiple actively and simultaneously executing copies of various operating systems (OS) to implement separate corresponding individual and unique application processing environments. (Enterprise System/9000 is a registered trademark, and ES/9000 is a trademark, of IBM Corporation.) Each of these environments utilizes a separate copy of the operating system, such as the MVS/ESA (henceforth simply “MVS”) OS, which forms a so-called OS “image”, along with an instances) of corresponding application program(s) and a dedicated storage area (typically a logical partition—“LPAR”). (MVS/ESA is a trademark, and IBM is a registered trademark, of IBM Corporation) As such, each such environment thus constitutes a separate “processing system” (henceforth referred to, for the sake of brevity, as simply a “system”). Each application instance that executes on any such system constitutes a separate application server (henceforth referred to as simply a “server” or “real instance”) to service a portion of the total workload presented to the overall application on the sysplex. A system, based on its processing capacity and that required by the corresponding applications, can implement one or more corresponding servers.
A recurring difficulty in using multiple servers has been how to effectively balance the current processing load across the servers. Traditionally, operating systems, such as the MVS OS, relied on a totally static approach to allocating available sysplex resources, such as available servers, processing time, and-processor storage, to each current work request. To accomplish this, system administrators utilized historic performance measurements of past workload processing to project just what sysplex resources would then be available as each new work request was presented to the sysplex and how these available resources should be allocated to handle that request. The overall goal of the administrator in allocating these resources to the current work requests was simply to keep each system maximally busy, i.e., to utilize as many available clock cycles thereon as possible, in effect keeping that system “pegged” and hence maximizing its throughput.
For a sysplex, historic averaged performance measurements were made over a variety of intervals and in relation to a variety of causes: e.g., on a day-by-day basis, on an hour-by-hour basis, and by each individual application, as well as in relation to other time or usage-related metrics. Based on this data, a stem administrator determined, from projections made from this historic data: how current work requests should be assigned to individual servers, a dispatching priority for each one of these requests that would be queued on each server, i.e., the order in which these requests were to be executed on that server, and the amount of resources at that server to allocate to each new work request presented thereto. Once these determinations were made for an expected sysplex workload in view of the goal of maximizing throughput of each system, the administrator simply instructed the operating system at each server accordingly. Through this effort, the administrator strove to distribute the total workload, as he or she then foresaw it, across all the servers as evenly as possible consistent with maximizing the throughput of all the servers.
Unfortunately, dispatching relationships existing between different work requests queued for execution in a sysplex tend to be extremely complex. Not only were accurate predictions of workload and resource allocations across multiple servers extremely tedious and difficult to create, but also such allocations were based on static, i.e., fixed, workloads having concomitant demands for each server that were not expected to change over time. Unfortunately, in practice, workloads do change, often significantly with time. Predictions predicated on static workloads simply could not accommodate subsequent changes in sysplex workload. Hence, each time a new application or a change in arrival patterns or demand for existing workload was to occur in a sysp
Aman Jeffrey David
Emmes David Bruce
Palmieri David Walsh
Donaghue Larry D.
International Business Machines Corporartion
Kinnaman, Jr. William A.
Michaelson Peter L.
Michaelson & Wallace
LandOfFree
Apparatus and accompanying method for assigning session... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus and accompanying method for assigning session..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus and accompanying method for assigning session... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2475422