Method and system for load balancing in a distributed object system

[0002] This invention relates generally to object-oriented programming and more particularly to a method and system for load balancing in a distributed object system.

[0003] When large-scale distributed object systems are built, scalability and performance are major concerns. Scalability refers to the ability of a system to scale up to support a larger number of users, a larger workload, a larger number of transactions, and/or a larger number of services. In other words, scalability refers to the ability of a system to easily expand to perform more work. A system with good scalability allows one to introduce more resources as needed to scale the system to support the additional workload without making significant changes to the remainder of the system.

[0004] In typical distributed system environments, the load on the system changes, sometimes on a monthly basis and even on a daily basis. A flexible system should ideally make efficient use of all resources in the system to balance the load across the network and achieve high performance.

[0005] Software developers have begun considering the implementation of large distributed object systems. In such systems, it may be desirable for the systems manager to balance the workload of various computer processes across the network. Accordingly, a need has arisen for a scalable distributed object system that attains high performance by efficiently distributing the workload across a network.

[0006] The invention provides a method and system for load balancing in a distributed object system that facilitates the scalability and efficient performance of that system. One aspect of the invention is a method for load balancing in a distributed object system running on a network comprising a plurality of computers including a first computer wherein the computers are operable to access a plurality of shared objects in a distributed object system. The method comprises instructing an object comprising a part of an application process running on the first computer to record at least one performance statistic in response to a message directed to the object. The application process comprises a multi-threaded process including a statistics thread. The at least one performance statistic may be periodically obtained using the statistics thread and sent to a local agent process running on the first computer. The performance statistic is related to a workload service running on a second computer connected to the network. A new distributed object is instantiated in the memory of one of the plurality of computers based upon performance statistics maintained by the workload service.

[0007] The invention has several important technical advantages. By using a workload service to determine where new objects are instantiated, the invention allows efficient use of all computer resources in a distributed object system. The disclosed system is highly scalable. Because the invention takes into account the performance of the system in determining where to instantiate new objects, the system makes efficient use of resources to achieve higher throughput. The architecture of the system allows each business object (or application object) to maintain its own statistics, thus allowing the workload service and local agents to operate smoothly without regard to how many application processes are providing the workload service and local agent with statistics.

[0008] For a more complete understanding of the present invention and the advantages thereof, reference is now made to the following descriptions, taken in conjunction with the accompanying drawings, in which:

[0009] FIG. 1 illustrates an exemplary general purpose computer that may be used to implement the present invention; and

[0010] FIG. 2 illustrates an exemplary distributed object system constructed in accordance with the invention.

[0011] The preferred embodiment of the present invention and its advantages are best understood by referring to FIGS. 1-2 of the drawings, like numerals being used for like and corresponding parts of the various drawings.

[0012] FIG. 1 illustrates a general purpose computer 22 that may be used in a distributed object system created in accordance with the invention. The general purpose computer 22 may be used to execute distributed application processes and/or distributed system services in accordance with the invention. General purpose computer 22 may be adapted to execute any of the well-known OS2, UNIX, MAC-OS and WINDOWS operating systems or other operating systems. General purpose computer 22 comprises processor 24, random access memory (RAM) 26, read only memory (ROM) 28, mouse 30, keyboard 32 and input/output devices such as disk drives 34, printer 36, display 38 and communications link 40. The present invention includes programs that may be stored in RAM 26, ROM 28 or disk drives 34 and may be executed by processor 24. Communications link 40 connects to a computer network but could be connected to a telephone line, an antenna, a gateway or any other type of communications link. Disk drives 34 may include a variety of types of storage media such as, for example, floppy disk drives, hard disk drives, CD-ROM drives or magnetic tape drives. Although this embodiment employs a plurality of disk drives 34, a single disk drive 34 could be used without departing from the scope of the invention. FIG. 1 only provides one example of a computer that may be used with the invention. The invention could be used on computers other than general purpose computers, as well as on general purpose computers without conventional operating systems.

[0013] FIG. 2 illustrates an example of a distributed object load balancing system 41 constructed in accordance with the invention. Distributed object load balancing system 41 comprises two or more computers 42, 44 networked together. Each of the computers 42, 44 may be similar in structure and operation to general purpose computer 22 illustrated in FIG. 1. Other types of computers could also be used without departing from the scope of the invention. Mainframe computers and mini-computers as well as personal computers (also known as microcomputers) may be useful for implementing distributed object systems in accordance with the invention. Each of the computers 42, 44 may be networked to one another in any way that multiple computers may be connected.

[0014] Distributed object load balancing system 41 comprises computer 42 with workload service software 46 running on it. Workload service 46 is operable to receive performance statistics for various application processes running in the distributed object system. The term, distributed object system, as used in this application is meant to refer broadly to any system that uses distributed objects. Workload service 46 may then be used to determine which of the plurality of computers 42, 44 in the distributed object system should be used to create a new object in the memory of one of the computers 42, 44 where the new object comprises a part of the distributed object system. Although workload service 46 could be designed with many different goals in mind, workload service 46 preferably causes objects in a distributed object system to be created in the memory of one of the computers 42, 44 in such a way as to balance the workload of each of the computers 42, 44. The invention thus allows a distributed object system constructed in accordance with the invention to adapt to varying traffic patterns in the system and make efficient use of various available resources.

[0015] The workload service 46 collects detailed statistical information about what different objects are active in different processes, the computer 42, 44 on which the objects reside, what methods have been invoked on them, how many times those methods have been invoked, and how much time has been spent in executing the methods. In one embodiment, the amount of time spent for a given method is measured in terms of CPU time but could be measured by other types of time measurements. The present invention allows statistics gathering for the aggregate of all objects of a particular type. For example, a bank attempting to keep track of customer accounts in a distributed object system may have a customer object. Suppose that the bank has only three customers A, B & C. Workload service 46 may maintain the statistics for objects A, B and C, separately, and may also maintain cumulative numbers for all customers. Thus, workload service 46 may maintain the above-described statistics for all customer objects cumulatively as well as other statistics on a cumulative basis. Average statistics for all instances of a given object class might also be maintained by workload service 46. Thus, when workload service 46 attempts to determine which computer 42, 44 should contain a distributed object in its memory, the workload service 46 may take into account not only the present workload of the computers 42, 44 in the distributed object system, but may also make the decision based upon a prediction of the resources that will be consumed by the new object due to the average workload that objects of that particular object class have previously imposed on computers 42, 44. Note that more statistics or less statistics could be collected by workload service 46 without departing from the scope of the invention.

[0016] Because the application processes 50 running on various computers 42, 44 may be performing critical tasks, the invention seeks to perform valuable statistics gathering without interfering with the operation of these application processes 50. Note that an actual application may comprise many application processes 50. The invention achieves this goal by providing a statistics thread 54 within each application process 50 to be responsible for gathering statistics. Statistics thread 54 may be transparent to an application developer who is developing a distributed object system application process 50. A distributed object framework may be provided to an application developer such that statistics thread 54 is automatically incorporated into an application process 50 when the application developer chooses to use that framework. Statistics thread 54 avoids interfering with the function of application process 50 by running asynchronously and avoiding interruption of the actual business tasks being conducted by application process 50. In this example, application process 50 is a multi-threaded application process comprising a statistics thread 54 and a main thread 52. Main thread 52 and any other application threads may be used to perform whatever task for which application process 50 is designed to perform. Additional threads could be included without departing from the scope of the invention.

[0017] In this embodiment, an interceptor thread (not explicitly shown) for each application process 50 intercepts messages intended for objects of application process 50. The interceptor thread is responsible for informing an object to update its performance statistics upon completion of a given operation. Thus, application process 50 instructs its own objects to gather performance statistics regarding themselves. Each object has access to a statistics data structure in memory of the computer 42, 44 on which it is running. The statistics data structure for a given application process 50 resides in the memory space for that application process 50. When an object completes an operation, it updates the statistics data structure with the statistics described above in connection with workload service 46. Additional statistics could be recorded or some of the above statistics excluded without departing from the scope of the invention. One option for avoiding interference with the operation of application process 50 is to only maintain performance statistics for messages received by application process 50 that originated outside of the application process 50.

[0018] Statistics thread 54 may be configured by the developer of application process 50 to periodically wake up, gather the statistics on objects currently residing in the memory of the computer 42, 44 on which the application process 50 is executing, and send these statistics to workload service 46 through local agent 48. Statistics thread 54 may be programmed to send either an empty message or no message at all to local agent 48 if no new statistics have been generated since the last time statistics thread 54 woke up. Such an action may be considered to be part of the process of waking up and forwarding of statistics. Thus, when this application refers to periodically waking up and obtaining performance statistics, that action encompasses obtaining no information during some of the periods. Eventually, the statistics thread will obtain a performance statistic during one of the periodic wake up times. The statistics thread gathers statistics on objects residing in the memory space of the application process 50 with which it is associated. In this embodiment, statistics thread 54 wakes up periodically, accesses the data structure containing the statistical data and sends the statistical information to local agent 48. Statistics thread 54 then goes to sleep. The time between the periods of statistics gathering by statistics gathering thread 54 may be adjustable either during development of application process 50 or by a system administrator during use of application process 50.

[0019] Local agent 48 receives performance statistics from various statistics threads 54 and relays those statistics to workload service 46. The invention thus avoids interference with application process 50. In a distributed object network environment, no assumptions can be made about the speed of the network and the availability of various services on the network. In addition, statistics data is eventually reported from many different local agents 48 to a central workload service 46. Because the workload service 46 may be busy receiving data from several local agents 48, it may delay the reporting of data from other local agents 48. Because local agents 48 receive their data from statistics threads 48 resident on the same computer 42, 44, local agent 48 may receive the statistical data immediately from the statistics thread 48, freeing up the application process 50 to continue performing its function. Local agent 48 may save the statistics data on a persistent storage medium and relay it to workload service 46 when the network is not busy or when workload service 46 is ready to receive the data. In an alternative embodiment, statistics thread 48 could perform the functions of local agent 48 such as forwarding statistics to the workload service.

[0020] Although the operation of distributed object load balancing system 41 has been described above, it will now be briefly summarized for an example embodiment. Each application process 50 maintains performance statistics regarding its objects resident in memory of the computer 42, 44 that is running the application process 50. Periodically, statistics thread 54 wakes up and relays those statistics to local agent 48. Local agent 48 relays the performance statistics to workload service 46. When it is desired to instantiate a new application object, the decision of which application process 50 is to instantiate and contain the new application object is based upon performance statistics maintained by workload service 46. Any suitable formula or algorithm may be used for this determination.

[0021] Although the present invention has been described in detail, it should be understood that various changes, substitutions, and alterations can be made hereto without departing from the spirit and scope of the invention as defined by the appended claims.