IBM Corporation. The IBM Interactive Network Dispatcher, 1998. http://www.ics.raleigh.ibm. com/netdispatch .  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and a cluster of servers. The emphasis of the MagicRouter work is on reducing packet processing time through Fast Packet Interposing but not on the issue of balancing load. Other solutions based on similar architectures include Cisco s Local Director [7] and IBM s Interactive Network Dispatcher [16]. An architecture slightly different from that of the MagicRouter is described in [4] in which a TCP Router acts as a front end that forwards requests for Web service to the individual back end servers of the cluster. Two features of the TCP Router differentiate it from the MagicRouter solution ....

IBM Corporation. "The IBM Interactive Network Dispatcher" at Http://www.ics.raleigh.ibm.com/netdispatch

....cluster reserves are being used. With cluster reserves, all service classes achieve their allocated resource usage. 6. RELATED WORK Network servers based on clusters of workstations are now widely used [14] Several commercial products are available for use as front end nodes in cluster servers [11, 16]. 5 10 15 20 25 30 35 0 20 40 60 80 100 service1 service2 service3 CPU usage ( Time (secs) Figure 15: With Cluster Reserves Most request distribution strategies employed are variations of WRR (described in Appendix A.1) The LARD [20, 4, 5] strategy is based on content based ....

IBM Corporation. IBM interactive network dispatcher. http://www.ics.raleigh.ibm.com/ics/isslearn.htm.

....the purpose of measuring webserver performance representative of real servers. Our methodology is also applicable to other operating systems and for measuring the capacity of faster servers, whether based on a single machine (like Flash[14] and Zeus[15] or on a cluster (like IBM s Netdispatcher[16] and LARD[17, 18] The key for making this possible is to realize that the 4 tuple (client IP address, server IP address, client port number, server port number) should not repeat itself for a period of time as long as the wait in the TIME WAIT period. For this purpose, we changed our clients as ....

IBM Corporation, "IBM interactive network dispatcher," http://www.ics.raleigh.ibm.com/ics/isslearn.htm.

....One common feature of all of these attempts (whether proposed or implemented) is that a centralized mechanism is employed to perform the mapping from IP addresses to hosts. Examples include the Berkeley MagicRouter [2] the Cisco Local Director [17] and IBM s TCP Router [6] and Network Dispatcher [4]. In contrast, DPR is a technique that allows the mapping between IP address and host to be implemented in a distributed, efficient, and scalable fashion. In particular, DPR can be viewed as a distributed method of mapping m IP addresses to n servers. If m = 1, then DPR becomes similar to the ....

....to the total capacity of its constituent servers. Thus, solutions that impose a large overhead are not desired. 4. Graceful Degradation: The failure of a system component should result in a proportional degradation 1 If m = 1, the difference between DPR and the centralized solutions proposed in [2, 6, 4] is that DPR allows both packet routing and service to be combined on the same node. in the offered quality of service. For example, a solution that allows for a single point of failure may result in major disruptions due to the failure of a miniscule fraction of the system. 5. Connection ....

[Article contains additional citation context not shown here]

IBM Corporation. The IBM Interactive Network Dispatcher. See http://www.ics.raleigh.ibm. com/netdispatch.

....node is capable of serving any target, although in certain request distribution strategies, the front end may direct a request only to a subset of the back ends. 2. 2 Aiming for Balanced Load In state of the art cluster servers, the front end uses weighted round robin request distribution [7, 14]. The incoming requests are distributed in round robin fashion, weighted by some measure of the load on the different back ends. For instance, the CPU and disk utilization, or the number of open connections in each backend may be used as an estimate of the load. This strategy produces good load ....

....like TCP. The front end must establish a connection with the client to inspect the target content of a request prior to assigning the connection to a back end node. The established connection must then be handed to the chosen back end node. State of the art commercial cluster front ends (e.g. [7, 14]) assign requests without regard to the requested content and can therefore forward client requests to a back end node prior to establishing a connection with the client. Our handoff protocol is transparent to clients and also to the server applications running on the back end nodes. That is, no ....

[Article contains additional citation context not shown here]

IBM Corporation. IBM interactive network dispatcher. http://www.ics.raleigh.ibm.com/ics/isslearn.htm.

....in its internal state that allows it to forward subsequent packets for that connection to the same server. The dispatcher must process all external client packets entering the cluster, but need not see packets from the server to the client. Several announced products use this or similar methods [8, 10, 13, 18, 25], which is illustrated in Figure 1. IBM s product is used for the main IBM web site (www.ibm.com) and was used for the Summer 1996 Olympic Web Server [3, 7] In IBM s implementation,the network dispatcher includes a manager which is used to implement load balancing policies. The network dispatcher ....

IBM Corporation. IBM interactive network dispatcher. http://www.ics.raleigh.ibm.com/ics/isslearn.htm.

....One common feature of all of these attempts (whether proposed or implemented) is that a centralized mechanism is employed to perform the mapping from IP addresses to hosts. Examples include the Berkeley MagicRouter [2] the Cisco Local Director [17] and IBM s TCP Router [7] and Network Dispatcher [9]. Distributed Connection Routing using DPR: In contrast, DPR is a technique that allows the mapping between IP address and host to be implemented in a distributed, efficient, and scalable fashion. In particular, DPR can be viewed as a distributed method of mapping m IP addresses to n servers. 1 ....

....and connection routing capacities. To illustrate the benefits of using DPR, consider the problem of scaling up a Web site that initially consists of a single server host. Adding a second server host using typical existing solutions (for example, Cisco s Local Director [17] or IBM s NetDispatcher [9]) requires using special purpose hardware to distribute incoming HTTP requests between the two server hosts. This kind of centralized solution provides connection routing capacity that far surpasses what a twohost server is likely to require. In other words, the upgrade path (and hence the price ....

[Article contains additional citation context not shown here]

IBM Corporation. The IBM Interactive Network Dispatcher. See http://www.ics.raleigh.ibm.com/ netdispatch.

....in cluster based network servers, where the front end takes into account the service content requested when deciding which backend node should serve a given request. In contrast, the purely load based schemes like weighted roundrobin (WRR) used in commercial high performance cluster servers [15, 8] distribute incoming requests in a round robin fashion, weighted by some measure of load on the different back end nodes. The potential advantages of content based request distribution are: 1) increased performance due to improved hit rates in the back end s main memory caches, 2) increased ....

....the complementary issue of providing support for HTTP 1.1 in cost effective, scalable network servers. Network servers based on clusters of workstations are starting to be widely used [12] Several products are available or have been announced for use as front end nodes in such cluster servers [8, 15]. To the best of our knowledge, the request distribution strategies used in the cluster front ends are all variations of weighted round robin, and do not take into account a request s target content. An exception is the Dispatch product by Resonate, Inc. which supports content based request ....

[Article contains additional citation context not shown here]

IBM Corporation. IBM interactive network dispatcher. http://www.ics.raleigh.ibm.com/ics/ isslearn.htm.

No context found.

IBM Corporation. The IBM Interactive Network Dispatcher, 1998. http://www.ics.raleigh.ibm. com/netdispatch .

No context found.

IBM Corporation. "The IBM Interactive Network Dispatcher". See Http://www.ics.raleigh.ibm.com/netdispatch

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC