Electrical computers and digital processing systems: multicomput – Computer network managing – Network resource allocating
Reexamination Certificate
2000-02-25
2003-12-02
Jaroenchonwanit, Bunjob (Department: 2141)
Electrical computers and digital processing systems: multicomput
Computer network managing
Network resource allocating
C709S229000
Reexamination Certificate
active
06658473
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to the field of networked computer systems in a multiple server environment.
2. Background Art
Computer users continue to desire high performance computing experiences in ever-changing computer environments. The computing paradigm is shifting. New architectures are emerging which require new solutions to deal with the need for a high performance computing experience. One such architecture is that of the thin client computing system. In this architecture, the functionality of the end user computer is reduced to the point that, for the most part, only input and output capabilities exist. The end user computer is connected over a high bandwidth computer network to a more powerful server computer which performs all the functions traditionally associated with the personal computer, such as executing computer programs and processing data.
In this type of architecture, a large number of end users can connect to a limited number of servers. In addition, the limited number of servers are also interconnected, creating what is termed as a multiple server environment wherein any of the end user terminals could potentially connect to any of the servers. In multiple server environments it is common for the environment to be heterogeneous, in that each server has differing resource capabilities. In such complex multiple server environments, the load on the servers' resources often becomes unbalanced, meaning, for example, that one server is performing at essentially maximum capacity while another server is relatively unused. Overcoming this load imbalance, therefore, becomes an extremely important concern, if a high performance computing experience is to be provided.
The evolution that led to this problem is better understood by reviewing the development of network computing. The rise of the internet has resulted in the proposed use of so-called “network computers.” A network computer is a stripped down version of a personal computer with less storage space, less memory, and often less computational power. The idea is that network computers will access data and applications through a computer network, such as the internet, intranet, local area network, or wide area network. Only those applications that are needed for a particular task will be provided to the network computer. When the applications are no longer being used, they are not stored on the network computer.
Recently, a new computer system architecture referred to as the virtual desktop architecture has emerged. This system provides for a re-partitioning of functionality between a central server installation and the user hardware. Data and computational functionality are provided by data sources via a centralized processing arrangement. At the user end, all functionality is substantially eliminated except that which generates output to the user (e.g. display and speakers), takes input from the user (e.g. mouse and keyboard) or other peripherals that the user may interact with (e.g. scanners, cameras, removable storage, etc.)
All computing is done by one or more servers acting as central data sources and the computation is done independently of the destination of the data being generated. The output of a data source is provided to a terminal, referred to herein as a “Desktop Unit” (DTU). The DTU is capable of receiving the data and displaying the data.
The virtual desktop system architecture may be analogized to other highly-partitioned systems. For example, a public telephone company maintains powerful and sophisticated processing power and large databases at central offices. However, the DTU, (e.g., the telephone), is relatively simple and does not require upgrading when new features or services are added by the telephone company. The telephone itself becomes an appliance of low cost and extremely low obsolescence. Similarly, the display monitor of most computer systems has low obsolescence, and is typically retained through most desktop system upgrades.
The provision of services in the virtual desktop system architecture revolves around an abstraction referred to herein as a “session.” A session is a representation of those services which are executing on behalf of a user at any point in time. The session abstraction is maintained by facilities known as the authentication and session managers, whose duty it is to maintain the database of mappings between tokens (i.e., unique identifiers bound to smart cards or other authentication mechanisms) and sessions, and to manage the services which make up each session. For each user that the system is aware of there are one or more sessions. The session manager offers a service to the user that allows sessions to be configured and new sessions to be created.
In a multiple server environment, multiple sessions may be executing on each server. These sessions are initiated by multiple users accessing the DTUs. If one of these servers fails (e.g., loses power), each of the DTUs connected to it “fails over” to one of the surviving servers. Since the computational and memory resources allocated to the services requested by the DTUs are distributed across the group of servers, it is possible for resources to become unevenly allocated, thereby degrading performance on over-utilized servers while wasting resources on under-utilized servers. This is especially true in heterogeneous server configurations, where the carrying capacity of the servers (i.e., number and speed of processing units, amount of installed memory and available network bandwidth, for instance) is non-uniform. In addition, each session may demand differing quantities of resources adding to the non-uniformity of the resources allocated.
Furthermore, the presence of failures complicates the load distribution problem, because if a server hosting a large number of DTUs fails, all of the DTUs will attempt to fail over within a short time period. It is crucial in this situation not to unbalance the remaining servers by connecting failed over sessions to a single server or an already overburdened server. Clearly, a more intelligent load balancing strategy is needed to achieve optimal resource allocation in this complex multiple server environment.
SUMMARY OF THE INVENTION
The present invention provides a method and apparatus for distributing load in a multiple server computer environment. In one embodiment, a group manager process on each server periodically determines the server's capacity and load (i.e., utilization) with respect to multiple resources. The capacity and load information is broadcast to the other servers in the group, so that each server has a global view of every server's capacity and current load.
When a user attempts to access a DTU, the user inserts an identifier into the DTU which contains a unique token. This identifier is a smart card, in one embodiment. Once the identifier is inserted, the DTU uses the token to attempt to establish communications with the servers to start or resume one or more sessions. When a given DTU successfully starts or resumes a given session, the group manager process of that server first determines whether one of the servers in the group already is hosting the session for that token. If that is the case, one embodiment redirects the DTU to that server and the load-balancing strategy is not employed. Otherwise, for each resource and server, the proper load balancing strategies are performed to identify which server is best able to handle that particular session.
The load balancing strategies are designed to take into account one or more factors, such as the number and speed of the microprocessors at a given server, the amount of random access memory (“RAM”) at a given server, the amount of network bandwidth available to a given server, the number of sessions running on a given server relative to that server's carrying capacity (e.g., the maximum number of sessions that server can host), the states of sessions running on a server (e.g., active or inactive), and the expected usage habits of cer
Block Robert J.
Hanko James G.
Peacock J. Kent
Jaroenchonwanit Bunjob
O'Melveny & Myers LLP
Sun Microsystems Inc.
LandOfFree
Method and apparatus for distributing load in a computer... 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 distributing load in a computer..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for distributing load in a computer... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3182041