Multiplex communications – Data flow congestion prevention or control – Control of data admission to the network
Reexamination Certificate
1998-09-21
2001-06-05
Hsu, Alpus H. (Department: 2662)
Multiplex communications
Data flow congestion prevention or control
Control of data admission to the network
C370S392000, C370S423000, C370S463000, C709S250000
Reexamination Certificate
active
06243360
ABSTRACT:
BACKGROUND OF INVENTION
1. Field of the Invention
This invention relates to communication systems. More particularly, the invention relates to a server-workstation network with dynamic load balancing of messages in both the server inbound and outbound directions.
2. Related Patent Application
This patent application is related to copending U.S. patent application Ser. No. 08/297,469, filed Aug. 29, 1994, now abandoned which is a U.S. counterpart application of UK Patent Application GB 2281793, published Mar. 15, 1995 and filed Nov. 9, 1993 by G. D. Wallis et al and assigned to the same assignee as that of the present invention.
3. Description of the Prior Art
Routing, bridging and switching are existing techniques to optimize performance between workstations and servers. Recently, load balancing software has been introduced to further improve performance between work stations and servers. Load balancing software allows you to place multiple Network Interface Cards (NICs) into a server and attach the cards to one logical network. In addition, multiple balanced networks can be connected to one server. Normally, under load balancing software or other operating systems, a system administrator would assign a unique address to each NIC. With load balancing software, the NICs can be bound to one logical address, providing load sharing (balancing) across the server bus.
In operation, load balancing software uses dynamic balancing for outgoing traffic (data from the server to the workstation) and static balancing for incoming traffic (from the workstation to the server).
Dynamic balancing works by sending packets to each NIC in turn. For example, with four NICs in the server, the first packet is routed to NIC number 1; the second packet to NIC number 2; the third packet is routed to NIC number 3; the fourth packet to NIC number four; such that each NIC transmits every fourth packet.
Static balancing works differently. When a workstation wishes to connect to a server, it finds the Media Access Control (MAC) layer destination address of the server by sending a request, for example, an Address Resolution Protocol (ARP) on to the network. The address returned by the server in response to the ARP request is used by that workstation until further notice. The software assigns response addresses in round-robin fashion. NIC number 1 is assigned to workstations
1
,
5
,
9
and
13
; NIC number 2 is assigned to workstations
2
,
6
,
10
and
14
; NIC number 3 is assigned to workstations
3
,
7
,
11
and
15
and so forth. If, for example, workstations
1
,
5
and
9
all transmit essentially simultaneously (with static balancing) the server NICs will become unbalanced. That is, NIC number 1 will be overloaded while NICs 2-4 will be idle. As a result, the latency for each workstation will increase and the throughput on the network will decrease by the unbalanced loading of the NICs.
The present invention is directed to improving both the latency and the throughput on a network having a plurality of workstations coupled through a network switch to a server having multiple entry ports.
The prior art addressing load balancing on related networks is as follows:
U.S. Pat. No. 5,031,089 issued Dec. 30, 1988, describes a dynamic resource allocation scheme for distributed heterogeneous computer systems. A decentralized resource sharing scheme is distributed over several compute engines in the network. The disclosure requires a computation of a “workload value” and the transmission thereof to all nodes on the network.
U.S. Pat. No. 5,025,491 issued Oct. 26, 1989, describes dynamic address binding in communication networks. '491 patent describes the relocation of servers, for example in the event of a “crash”, and the rerouting of messages to a new server.
U.S. Pat. No. 5,271,003 issued Mar. 9, 1990, describes an internal routing method for load balancing. The '003 patent relates to rerouting messages in the event of a failed “message handling processor” in “circuit switched” networks.
U.S. Pat. No. 5,193,151 issued Aug. 30, 1989, describes a delay based congestion avoidance in computer networks. The '151 patent discloses a method to reduce traffic load based on a metric such as round-trip delay for a message and its associated acknowledgement.
U.S. Pat. No. 5,249,290 issued Feb. 22, 1991, describes a client/server computer network where the server uses a busy indicator to assign a new client service request to the least busy processor. The '290 patent uses a workload metric to assign “client” server requests to “servers”.
U.S. Pat. No. 5,293,488 describes a message routing apparatus that uses a Network Interface Card for each network and a shared routing manager. The '488 patent relates to “filtering” messages or frames in a routing environment to decide whether or not to forward the message or frame to a particular network.
U.S. Pat. No. 5,241,677 describes a multi-processor system and a method of load balancing. The '677 patent balances the computational load on processors to maximize multi-processing efficiency.
IBM Technical Disclosure Bulletin, Volume 37, Number 07, published July 1994, describes a remote log-in program to balance processing load. The publication discloses assigning a “client” to the least loaded “server” as measured by a non-specific metric.
None of the prior art describe a communication network including a network server with multiple Network Interface Cards coupled to a plurality of workstations through a network switch, the in-bound and out-bound messages to/from the server being dynamically balanced to improve latency, throughput and recovery from fault conditions on the network.
SUMMARY OF THE INVENTION
An object of the invention is to provide a communication network with improved latency and throughput.
Another object is to provide a communication network including multi-port server which always returns to a workstation the address of the first port installed in the server.
Another object is to provide a multi-port server having dynamically balanced network interface cards for in-bound and out-bound messages from/to a plurality of workstations.
Another object is to provide a multi-port server including a plurality of Network Interface Cards which are made “route equivalent” for inbound messages.
Another object is to provide a network switch which dynamically routes in bound messages to a first available port in a multiple port server.
Another object is to provide a network switch which matches a designated address for a data packet to a particular switch port.
Another object is to provide a method of dynamically balancing inbound and outbound messages between a plurality of workstations and a multi-port server.
These and other objects, features and advantages are achieved in a communications network including a server having multiple entry ports, each port served by a channel coupled to a plurality of workstations through a network switch. Each workstation includes a Network Interface Card (NIC) which sends/receives data packets to/from the switch after obtaining a Media Access Control (MAC) address of a NIC in the server with which it wishes to communicate. Each workstation NIC transmitting a packet to the server receives in return an address of the first entry port into the server or other network device that has more than one network interface card installed. The MAC address is stored by the workstation for future data packet transmissions. The workstation is connected to an associated input port of the network switch. The switch includes a destination port connected to each server port through a channel. The switch includes means to prepend a routing header to a data frame or packet transmitted to the switch by a workstation, the header identifying a particular destination port according to the destination of the data frame. The switch associates a destination port number with a server NIC address using, for example, a content addressable memory and a processor with a look-up table. The switch includes means to move the packet from the
Cockburn Joscelyn G.
Hsu Alpus H.
International Business Machines - Corporation
LandOfFree
Network server having dynamic load balancing of messages 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 Network server having dynamic load balancing of messages in..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Network server having dynamic load balancing of messages in... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2440091