WO2000019339A1

WO2000019339A1 - Selective connection flow redirection for transparent network devices

Info

Publication number: WO2000019339A1
Application number: PCT/US1999/022326
Authority: WO
Inventors: Anthony D. Amicangioli; Michael C. Shaw
Original assignee: Infolibria, Inc.
Priority date: 1998-09-28
Filing date: 1999-09-27
Publication date: 2000-04-06
Also published as: EP1125225A1; AU6401899A

Abstract

The present invention is a technique for implementing a content server such as a cache server together with a message redirector which off-loads connection processing functions. The message redirector performs a function of filtering traffic away from the content server so that the number of connections for which the content server can contribute utility is maximized. This is done by performing a load shedding functionality for accepting new connections and passing through new connection requests while the content is overloaded.

Description

SELECTIVE CONNECTION FLOW REDIRECTION FOR TRANSPARENT NETWORK DEVICES

BACKGROUND OF THE INVENTION

Computer network industry analysts and experts agree that data traffic over large networks and, in particular, the Internet, is presently so heavy that the very nature in the way in which it is possible to use such networks may require fundamental changes . These difficulties are no doubt the result of continued exponential increases in the number of users, as well as in the number of large document files that these users expect to transfer. As a result of this unprecedented demand in the need for network bandwidth, Internet Service Providers (ISPs) , backbone providers, and other carriers that provide the physical connections which implement the Internet face correspondingly unprecedented difficulty. This difficulty exists at all levels of network hierarchy, including the points of presence (POPs) , central access nodes, network access points, and exchange points, such as metropolitan area exchanges. To understand what has led to this situation, consider how the Internet is typically used. The most popular user environment, the World Wide Web, provides access to content on an equal basis through the use of the client and server communication model. In this structure, certain computers known as "servers" are used to store and provide information. Other computers in the network known as "clients" allow the users to view documents through the use of a computer program known as a browser that requests a copy of the document be sent from host servers down to the client. Documents are typically requested by the client browser program specifying an address which identifies the host server which stores the document . After the browser specifies a document such as by its so-called Internet Protocol (IP) address, the request is sent to a naming service in order to obtain instructions for how to establish a connection with the host server associated with the IP address. Once this connection is established, the server retrieves the document from its local disk and transmits the document over network to the client. The connection between the client and host server is then terminated. A common solution for the present bottlenecks within the Internet is to deploy higher speed hardware. Such solutions include the deployment of digital subscriber line (xDSL) cable modem technology to speed up the physical layer communication paths between the end users and points of presence. Gigabit speed routers and optical fiber backbones are also being proposed to alleviate congestion within the network itself. At the server site, server clusters and load balancers are being deployed to assist with the dispatching of Web pages more efficiently.

While all of these solutions provide some expediency, each addresses only part of the problem and none provides an ultimate solution to the problem; the path between the client and server is only as fast or as slow as the slowest physical link.

As it turns out, much of the traffic on the Internet is redundant in the sense that different users request the same documents from the same servers over and over again. Therefore, it is becoming increasingly apparent that certain techniques, such as distributed document caching, may be deployed to reduce the demand for access to both the servers and to the network routing infrastructure. Document caching provides a way to intercept client requests for the same document with the cache serving copies of the original document to multiple client locations. With a cache, the process for providing document files to the client computers changes from the normal process. In particular, when the client requests the connection, say to a given server, the intermediate cache server may instead be requested to obtain the document. While the document is being transmitted down to the client computer, a copy is stored at the intermediate cache server. Therefore, when another client computer connected to the same network path requests the same content as the first user, rather than requiring the request to travel all the way back to the host server, the request may be served from the local cache server.

However, cache techniques are typically sub-optimal in one way or another. For example, most Web browser programs have a built-in cache that keeps copies of recently viewed content within the client computer itself. If the same content is requested again, the browser simply retrieves it from local storage instead of going out to the network. However, when a browser cache services only one end user, content often expires before it can be reused. A browser-redirected cache server may also be deployed to service multiple end users. Such a cache server is a separate computer that sits inside a gateway or other point of presence . End users configure their Web browsers to redirect all HTTP traffic to the cache server instead of the locations implied by the Uniform Resource Locators (URLs) . The cache server then returns the requested Web page if it has a copy. Otherwise, it forwards the request to the originally specified server and saves a copy as the response flows back. Such a cache server therefore acts as a proxy, receiving all requests and examining them to determine if it can fulfill them locally. However, even when using proxy servers, it is typically necessary to configure the client browser, proxy server, routers, or other network infrastructure equipment in order to cause the request messages to be redirected to the proxy server. This provides some configuration management difficulties in that reconfiguration of browsers typically requires administrative overhead on the part of the humans who manage the networks .

To improve the odds of locating desired content without having to traverse the entire Internet, local points of presence can be supported by additional caches placed deeper into the network, such as at peering centers. If a primary cache cannot satisfy a request, it queries a secondary cache which in turn may query a tierciery cache, and so forth. If none of the caches in the hierarchy has the desired content, the primary cache ultimately ends up forwarding the document request to the originally requested host. These caching schemes also fall short in some way. Forced redirection of HTTP traffic turns such cache servers into single points of failure. If a cache server overloads or malfunctions, access to the network is blocked. Recovery is especially awkward with browser redirected caching since every end user's Web browser then has an explicit point to the broken server.

Forced redirection can also have a negative effect on network performance. Even if a browser is topologically closer to the real content server than to a cache server, all HTTP requests detour through the cache and any Web object not in the cache passes through the nearby router or switched twice — one when it travels from the originating server to the cache, and again as the cache forwards it back to the browser and furthermore passing messages from primary to secondary caches and back again at its noticeable latency and ultimately limits the scope of caching in larger networks.

Therefore, there is presently much controversy over the deployment of network caches for several reasons. Cache servers are, in particular, notoriously difficult to optimize. In certain configurations, they will quickly become overloaded in that the number of connections that they are expected to maintain with the user locations is more than the processing power can handle. Time spent determining whether to accept connections, cache documents, and/or refuse connections therefore overloads the cache server, which in turn reduces its performance on an exponential basis. In other situations, the cache servers are underloaded and not enough traffic is routed to them. They therefore represent a large investment of resources not providing optimum utilization and which are in effect underloaded.

SUMMARY OF THE INVENTION

The present invention is a technique for implementing a content server such as a cache server together with a message redirector which off-loads connection processing functions. The message redirector performs a function of filtering traffic away from the content server so that the number of connections for which the content server can contribute utility is maximized. This is done by performing a load shedding functionality for accepting new connections and passing through new connection requests while the content server is overloaded.

More particularly, the message redirector is a three-port, transparent bridge with enhanced features such as filtering and traffic redirection. The bridge function permits a cache server to be transparently installed in-line between routers, switches, and other network backbone infrastructure.

A connection selectivity function performed by the message redirector provides increased hit rate for the cache server as measured in the number of objects delivered from the cache versus the number of objects which must be retrieved from elsewhere in the network. This connection selectivity functionality is implemented by maintaining a list of objects that are stored in the cache, as has been done in the past. However, in the present invention, the list itself is kept in a distinct location outside of the cache server, in the redirector.

During the process of attempting to service each document request, the message redirector first looks for a new connection request. Such a request may take the form of a SYN packet. The Internet Protocol (IP) address associated with the new connection request is then compared to the local selective connection table in the message redirector. For a predetermined period of time, referred to as the selectivity period, the SYN request is only routed up to the cache server if it is in the selective connection table.

At the end of the selectivity period, SYN requests are again permitted to be routed to the cache server, whether or not they are located on the selective connection table. In this "promiscuous" mode, after the SYN has been routed to the cache server, the message redirector again returns to the selective mode. The cache manager may also control the length of the selectivity period. This is typically set to an initially high value and progressively shortened until a point is reached near where the maximum possible serviced connections are being experienced by the cache manager . In general, when the cache is performing best, it is actually overloaded in the sense that the number of offered connections is greater than the number of serviceable connections.

In addition, every so often a new selectivity connection table is generated by the cache server. To accomplish this, the cache server periodically scans this list of objects stored in the cache to identify a subset of the most requested objects. An object may be a domain such as a full IP address or may be sub-net mask address. This list of popular requested addresses is then sent down to the message redirector. Therefore, it can be understood that optimization of the load on the cache server can be achieved by breaking the pool of available connections into two groups (1) a group of selective connections derived from high probability of hit rate sites and (2) a smaller group of non-selective connections.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.

Fig. 1 is a diagram illustrating an overview of how a cache server and message redirector can be deployed in a computer network according to the invention.

Fig. 2 is an alternate arrangement showing how multiple cache servers and message redirectors can be deployed cooperatively.

Fig. 3 is a more detailed view of the transparent message redirector.

Figs. 4A, 4B, and 4C illustrate the format of certain data structures used by the redirector. Fig. 5 is a flow chart of the steps performed by the message redirector to support layer two (L2) bridge functionality.

Fig. 6 is a flow chart illlustrating how selectivity and connection tagging/tracking is implemented in the filter logic.

Fig. 7 is a flow chart for processing received messages in the filter logic.

Fig. 8 is a flow chart for closing a connection in the cache server.

Fig. 9 is a flow chart for closing a connection in the redirector.

Fig. 10 is a table of connection tagging objects.

Fig. 11 is a diagram illustrating how tags are appended or inserted into a network message.

Fig. 12 is a plot of an exemplary optimum selectivity period versus number of offered connections .

Fig. 13 is an exemplary selectivity function.

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT

Fig. 1 is a detailed block diagram of a computer network 100 and the manner in which a content server 150 may make use of a redirector 140 to achieve the advantages of the present invention. The content server, which may for example be a cache server 150, may be deployed at any of the number of places within the infrastructure of the network 100. It may be deployed at network access sites, such as points of presence (POPs) , an Internet service provider (ISP) ,

ISP peering points, interchange points in a large scale enterprise network, central offices in a local exchange carrier network, metropolitan area exchanges, or other points through which the network message traffic is concentrated.

In the illustrated embodiment, the cache server 150 is deployed at an intermediate point in the network 100. The cache server 150 is configured to cache Web pages on a local disk 155. As Web page requests travel from a hypertext transfer protocol (HTTP) client 110 through a first set of network connections 112 to a first set of connected routers 114-1 and 114-2, a redirector 140 and cache server 150 then cooperate to determine whether the Web page can be served from the local disk 155. If so, the Web pages are returned through the routers 114-1,114-2 through the network connections 112 to the client 110. However, if the Web page requests cannot be served by the local disk 155, the original request is then forwarded on to travel through a second set of routers 114-3, 114-4, through other network connections 118, to eventually arrive at the originally requested HTTP server 170.

The message redirector 140 consists of three logical ports ho , eO, and el. The network ports eO and el each provide a physical connection to the network 100 such as to local area network (LAN) or wide area network (WAN) connections provided to the routers 114. The network ports eO and el may, for example, be compliant with ethernet lOBaseT, 100BaseT, or other types of physical layer implementations, such as ATM, PPC/Sonet, frame relay, or other network protocols. Although in the illustrated embodiment the ports eO and el are shown respectively connected to router 114-2 and router 114-3, it should be understood that they may provide connections to other access devices, switches, servers, bridges, and the like.

The host port hO provides a connection for passing message traffic to and from the host cache server 150. This port hO may also typically use more tightly coupled to connect physical layer connections such as local area network connections or may also be sharing a computer bus structure as will be described in more detail below. The redirector 140 also contains a controller to set the position of switches in the cut through switch 130, to permit message traffic to be routed up to the cache server 150 or passed straight through the between the routers 114-2 and 114-3. The message redirector 140 is a type of message processor with certain functions that will be described in greater detail herein. In the case where the server 150 is a cache server for served cached Web content, the message redirector 140 and cache server 150 cooperate to provide a transparent HTTP object cache for the network 100. In particular, the network cache 150 monitors HTTP traffic flow between the routers 114 and stores copies of sufficiently popular Web pages. Subsequent requests for the stored Web pages, for example from the HTTP client 110, are then retrieved from the disk 155 local to the cache server 150 rather than from the network HTTP server 170. This results in a significant reduction in network line utilization and improves user response time by reducing the number of hops between the client 110 and originally requested server 170, and also by providing multiple sources for popular Web content. In addition, this advantage occurs in a transparent manner so that the cache server 150 may be introduced into the network without specific reconfiguration of the HTTP client 110 or HTTP server 170. By being designed to be a transparent device relative to Internet Protocol (IP) network state and topology, the routers 114 to the need to be reprogrammed, and either does the browser program being used by the HTTP client 110. In effect, for non-HTTP traffic, the redirector 140 behaves as a transparent Ethernet bridge.

It should be understood that various arrangements of multiple redirectors 140, host cache servers 150 and the connections they make to the network 100 are possible. For example, turning attention to Fig. 2, the host cache server 150 may actually consists of a number of hosts 250-0, 250-1, ..., 250-2 and associated storage devices cooperating to increase the performance of the overall cache server 150. In this case, the redirector 140 may physically be implemented as a number of three port devices 240-0, 240-1, ..., 240-2 connected in series. The series connected redirectors 240 thus each provide physical access to the network 100 for each respective host 250-0, 250-1, ..., 250-2 at a respective host port hO , hi , ... , h.2. It should be understood, however that other multi-processing arrangements are possible, such as having the redirectors 240 arranged in parallel to share a computer bus type interconnection and the like.

Returning attention now to Fig. 1, the redirector 140 performs certain critical filtering operations to optimize the routing of messages to the cache server 150. These functions present messages to the cache server 150 in a manner which optimizes its utilization.

A back pressure function provide a control over how many new connections will be attempted to be offered to the cache server 150 depending upon its reported load.

The selectivity function provides most of the rest of logic necessary for transparency. Its primary function is to bridge packets that the host server 150 is not processing straight from port eO out through to port el. Packets that are being processed by the cache server 150 are passed up the stack through the one or more host ports hO , hi, ..., h2. The selectivity feature is in effect used to attempt to "hijack" only those packets which have a high probability of being related to HTTP objects or documents connections which the cache server 150 has already stored or, in the case of a new connection request, is capable of servicing . A hardware block diagram of a message redirector 140 which implements these features is shown in Fig. 3. It consists of a pair of network interfaces which may be integrated circuit chips or cards (NI (C) s) 302-1, 302-2, each associated with particular router connection, respectively, eO or el. The NI (C) s 302 forward packets to an internal bus structure 304 through one or more circuit interfaces such as PC industry standard architecture (ISA) , extended ISA (EISA), or PCI bus interfaces. A redirector controller 340, layer two (L2) data path controller 350, and payload memory 360 are also connected to the internal bus structure 302. The redirector controller function 340 includes a central processing unit (CPU) 310, an associated memory 312, timer 314, and a selective connection logic 316. The L2 and data path control circuitry 350 is used to control how packets are routed to and from the various elements of the redirector 140. It comprises circuits that control the state of the internal bus 304, to allow packets to be moved from the eO port through to the el port, or from one of the eO or el ports up to the redirector logic 340 and/or packet payload memory 350. In a multiport implementation where there are more than one host port present, the redirector logic 340 may be replicated for each host port hO , hi, ..., h.2 , or the redirector logic 340 may control more than one host port.

The payload memory 350 is used as a temporary buffer to store the payload portions of a packet while the rest of the redirector logic 340 is determining where to route the packets . Where small physical size is desirable, the internal bus 304 may be an industry standard PCI bus and the NI (C) s 302 may be integrated circuit chips. This configuration may be the least expensive for volume manufacturing in the case of a single host port implementation. In other implementations, the NI (C) s 302 may be individual circuit cards, and the bus 304 an external bus. This "external box" configuration may be more desirable for servicing multiple hosts 150.

Figs. 4A, 4B, and 4C are more detailed views of various data structures used in the redirector 140.

Fig. 4A is a layer two (L2) routing table 410. The L2 routing table 410 permits the redirector 140 to perform certain standard learning bridge functions. It therefore contains fields related to known L2 addresses in the system 100, including a Media Access Layer (MAC) layer address field 411. The L2 routing table is typically implemented using known fast lookup techniques such as hashing functions.

The table 410 also contains information specific to the redirector 140, such as a port field 412, a static/dynamic state field 413, and age field 414. The purpose of each of these fields will be described in greater detail in connection with Figs 5 and 6. It is sufficient here to note that the port field 412 is a decoded bit field indicating to which port (eO, el, or hO) a packet having the associated MAC layer address should be routed, the static/dynamic state field 412 indicates where layer 3 (L3) filters should be bypassed, and the age field indicates how long the entry has been in the table 410.

Fig. 4B illustrates a connection tracking object table 420 which will be described in greater detail in connection with Figs. 7 through 11. It includes entries associated with active connections that the cache server 150 is presently servicing, including an Internet Protocol source (IPs) address field 421, an IP destination (IPd) field 422, a TCP source (Ts) field

423, and TCP destination (Td) field 424. A port number field 420 indicates information relating to how to route packets internal to the cache server 150.

Finally, Fig. 4C illustrates a selective address table 430. As will be understood shortly, this table is used to determine whether or not a packet is actually routed up to the cache server 150 based upon a number of conditions. The entries in this table 430 include at least an IP address field 431 and a mask 432. An optional rating field 433 may be used to support a feature known as weighted selectivity; a port number field 434 is used in implementations supporting more than one host port hO , hi, ... , h2.

Fig. 5 is a more detailed flow chart of certain operations performed by the redirector logic 340 to perform the selectivity and connection tagging functions.

In an initial state 510, a packet is first received. Moving next to state 512, the source address of the packet is examined.

If the layer two (L2) (MAC) address is present on the L2 table 410, then processing can proceed to the next state 516. If however the L2 address is not on the L2 table 410, then this address has not yet been learned by the redirector 140. In this instance, a state 514 is first entered in which the source address and port number are added to the L2 table 410, to implement learning bridge functionality.

Continuing in state 516, it is then determined if the destination L2 address is located on the L2 table.

When the L2 address is not present, the redirector 140 simply passes the packet to the opposing ethernet port

(L2 broadcast) in state 517.

State 518 is next entered in which the L2 destination address is used to determine a tentative L2 forwarding decision. This decision, referred to herein as the FD_L2 decision, is indicated by reading at least the port number field 412 by finding the associated destination address in the MAC address field 411. Whichever bits are set in this field indicate to the redirector logic 140 to which port, e.g., eO, el, or hO (or hi, ..., h2 if present) the packet might be routed to, e . g. , tentative routing decision. In state 520, if the static/dynamic bit is set to indicate a static address, then this indicates a type of packet which is intended not for network connected devices 110 or 170, but rather a "forus" management layer packet intended for the cache server host 150 itself. This bit may also be set in the case of a MAC layer broadcast address, as show in Fig. 4A.

In this instance, state 522 is entered in which the packet is forwarded using the FD_L2 decision.

Otherwise, processing continues with a state 610 as indicated in Fig. 6. In this state, an initial packet filtering function is performed such as by packet type. In the case of a cache server 150 serving Web pages, the TCP header port number is read in state 610 to determine if it is an HTTP packet. It should be understood that other types of packet filtering may be performed at this juncture.

In the case of an HTTP packet filter, if the packet received is not an HTTP packet, such as if the TCP header port number is not set equal to "80" , then the packet is simply forwarded, or bridged, in state 612 using the FD_L2 decision.

Thus, for example, if the packet was received in on interface eO from router 114-1, it is forwarded directly out in the direction of router 114-4 on port el. Similarly, if the packet was received from router 114-3 on interface el, it is routed straight out through interface eO towards router 114-2. If, however, the packet is an HTTP packet, then processing proceeds to a state 61 . One consideration at this point relating to HTTP selectivity is IP fragmentation, which occurs because HTTP packets are sometimes fragmented into multiple IP packets. In such an instance, they will at this point need to be reassembled before they can be passed up the stack (assuming, for example, that the TCP header is present only in the first packet) . In any event, in state 614, the packet is examined to determine if it indicates a new connection (or "flow") such as if it includes an SYN packet. Such SYN packets indicate the beginning of a request for a connection for an HTTP object. If the packet is not a SYN packet, then this packet relates to a connection which has already been set up. Processing proceeds to state 650 in which other attributes of the connection are examined to determine if and how the packet is to be handled. If the packet is an SYN packet, then a new TCP connection is being requested. In state 616 and the following states, a selective connectivity feature determines whether or not a new connection should actually be established with the cache server 150. State 616 specifically determines if a maximum number of connections are already being serviced by the cache server 150. If this is the case, then processing proceeds to state 618 where the packet is simply bridged out of the redirector using the FD_L2 tentative decision.

If the cache server 150 is still accepting new connections, then processing proceed to the selective connection state 620, where it is determined if the redirector is in a "selective" or "non-selective" mode.

The connection selectivity function is a feature which attempts to maintain a list of the IP addresses of the servers 170 that contain the most popular objects stored in the cache server 150. A selective connection table (SCT) generation process, executing as part of the cache server, is responsible for generating information to permit the redirector 140 to maintain the list referred to the selectivity connection table (SCT) 430. This selective connection table 430 allows the message redirector 140 to hunt for connection requests (SYNs) that have a higher probability of a hit in the cache server 150, given that their destination IP address already has content loaded in the cache server 150. This selectivity feature also allows the cache server 150 to effectively shift the optimum cache locality point because it allows the cache server 150 to effectively participate in the need to compare fewer IP addresses.

Sub-net masks 432 and/or complete IP addresses may be used as the addresses stored in the selective connection table 430. For example, certain popular Web sites, such as cnn.com or yahoo.com, typically have a number of pages associated with them that may rise to the level of being sufficient popular to be maintained in the cache 150. In this instance, rather than maintain the complete four-digit full IP address for each page, a sub-net information mask 432 may be provided to indicate more than one page from the site is stored in the cache. In the preferred embodiment, a selectivity policy can be implemented through an initial time-based test to implement a selectivity period. The selectivity period is determined by a single timer circuit 314 that is global to all selective connections. Briefly, if a select connection (an SYN with an IP address in the selective connection table 340) is not found before the timer expires, the selective connection state switches to a non-selective mode. In this non-selective mode, any occurring SYN will be permitted to be routed up to the cache .

Thus, in the selective mode, only SYN requests which already have their associated IP addresses and/or sub-net masks stored in the selective connection table 430 are permitted to be routed out to the cache server. In the non-selective mode, the next SYN will be routed up. In this mode, the system provides an N/K selective to non-selective behavior.

As shown in Fig. 6, the connection selectivity function can be provided from state 620 as follows. In state 620, the contents of the timer 314 is detected and used to determine if the selective mode should be entered. If the timer indicates that it is not time to enter the selective mode, then processing can exit from state 616 to prepare the new connection by tagging it in state 660.

If, however, in state 620 the timer indicates the selective mode is active, then the state 622 is entered to look up the IP address of the SYN request. In state 624, if this address is located in the selective connection table 430, then the new connection will not be permitted to be maintained. In this instance, the packet is forwarded out of the redirector 140 using the tentative L2 decision FD_L2 in state 626. The connection therefore will not be serviced locally. However, if the destination address is on the selective table 430, then processing will continue with state 650.

If redirector is not in selective mode in state 620 or if IP destination is on the selective address table, then a connection tracking object and associated tag is assigned in state 650.

Assigning a connection tracking object for each active connection between the cache server 150 and a client on the network serves to off-load data processing at the redirector 140. For example, when a connection is to be maintained between the cache server 150 and client 110, multiple messages are typically exchanged between the client 110 and cache server 150. Since the redirector 140 and cache server 150 service multiple connections or communication sessions simultaneously, inbound messages from multiple clients 110, therefore, must be analyzed to determine to which connection a corresponding message should be directed.

Rather than analyze a newly received message at the cache server 150, the process of analyzing the source and destination information of received data messages is done at the redirector 140. After determining the connection associated with a message received at the redirector 140, a corresponding tag for that connection is appended to or incorporated into the message. As mentioned, this shifts the burden of processing inbound messages from the cache server 150 on to the redirector 140. Fig. 11 is a diagram illustrating how a tag 1125 is either appended to an original message 1115, thus, forming a tagged message 1130. Alternatively, a tag is optionally overwritten in an Ether field of the original message 1115 to create an embedded tag within the tagged message 1120.

This process of assigning an index tag to a connection and appending the corresponding index tag number simplifies bit manipulation at the cache server 150 because the cache server 150 receiving a tagged message from the redirector 140 needs only to read the tag to determine the associated connection to which the message pertains. Otherwise, many bits of information such as the IP source and destination and TCP source and destination address of the received message would have to be analyzed at the cache server 150 to determine the corresponding connection.

Fig. 10 is an array of N connection tracking objects 1000 for maintaining information associated with a particular connection. During operation, a similar array of connection tracking objects 1000 is maintained by both the cache server 150 and redirector 140. Accordingly, this provides the redirector 140 and cache server 140 a shorthand way of communicating which connection a message pertains. For example, the easy-to-read tag of a message passed between redirector 140 and cache server 150 indicates the connection to which the message pertains.

Each tag number is an index number corresponding to the connection entry in the connection tracking object array 1000. For example, connection tag #1 is an index pointer for the first object entry in connection tracking object table 1000, connection tag #2 is an index pointer for the second object entry in connection tracking object table 1000, and so on for each of an array of N connection tag objects 1005. Referring again to state 650 in Fig. 6, a connection and corresponding connection tag object 1005 must be established for the newly received data message and related subsequent messages. This process involves assigning a free connection tracking object 1005 in the connection tracking object table 1000 for the new connection.

After a connection tracking object 1005 is assigned for a new connection, the information associated with the connection is stored in the new connection tracking object 1005 in state 655. For example, the IP and TCP source and destination address of the connection are stored in the connection tracking object 1005 so that the TCP and IP source and destination addresses of other received messages can be compared to those in the connection array 1000 to determine whether the message pertains to an active connection.

Additionally a status 1010 of the connection tracking object is maintained, signaling whether a connection is active or inactive. In the present case, the status 1010 of the newly created connection tracking object 1005 would be set active. Further, a message type, i.e., UDP, TCP ..., is stored for the connection. Finally, a port number stored in the connection tracking object 1005 identifies which cache server 150 a connection pertains in the event that the system includes multiple cache servers 22. Following the storage of appropriate connection data for the connection tracking object 1005 in state 655, the corresponding index is appended or incorporated in the network message. In the preferred embodiment, the tag is stored in the Ethernet field or link layer. Thereafter, in state 670, the newly tagged message is forwarded to cache server 150 and is processed based on network layer 3 information. It is common for failures to occur in any networking system. Therefore, active connections are monitored for activity or communication between the cache server 150 and clients 110. If the communication on a given connection is inactive for a predetermined time, the connection tracking object at both the redirector 140 and the cache server 150 are closed, i.e., set inactive, freeing resources for new connections. This grace time can depend on the availability of resources and present traffic through the redirector 140. For instance, when the redirector 140 is plagued with traffic and there are no resources to open new connections, the grace time for a presumed failed connection may be shorter since the resources are optimally used to service other requests.

If the message received by the redirector 140 in state 614 is not a new message pertaining to a new flow, i.e., it does not include a SYN message, then it is thereafter determined if the message pertains to a an active connection tracking object 1010 in state 675. This involves comparing the TCP and IP source and destination address of the received message with active connection tracking objects 1010 in the connection array 1000. If a match is found indicating that the message relates to a particular connection in state 675, the redirector 140 marks the message with the corresponding tag index number in state 685. The tagged message is then forwarded to the cache server 150 in state 688 based on layer 3 redirect routing. Alternatively, if there is no corresponding connection open for a particular message in state 675, the message is forwarded to the cache server 150 based on layer 2 routing in state 680. Fig. 7 is a flow chart illustrating how messages received from the redirector 140 are processed at the cache server 150. The process involves first receiving a message from the redirector 140 in state 705. If the received message in state 710 does not include a connection tag, the message is sent to the appropriate socket using standard Unix TCP/IP routing in state 715.

If there is a connection tag associated with the received message in state 710, the message is passed on to state 720 to determine whether the message includes a SYN message. If not, the tagged message is directed within the cache server 150 to the connection running on the TCP state machine corresponding with the tag in state 730. Again, the tag number is an index number to the proper TCP state machine or session connection corresponding with the tagged message.

If the message received from the redirector 140 includes an SYN message and a tag in state 720, this indicates that a new connection is to be opened for the requested object. The corresponding tag is the index number of the connection in the connection tracking object array 1000 to be established for future communication of a particular connection. A connection tracking object 1005 is created including the information as described in Fig. 10.

Both the message redirector 140 and cache server 150 both track a particular connection based upon the content, at least in part, of each message. As a result, both the array in the redirector 140 and cache server 150 mirror each other, i.e., both arrays include substantially identical information, supporting the harmonious processing of messages. After it is determined to which connection the message pertains, the message is then processed in state 730 on the appropriate TCP state machine in the cache server 150.

Fig. 8 is a flow chart illustrating the process associated with closing or maintaining a connection in the cache server 150. First, the cache server 150 determines in state 805 if an object request by a node is properly serviced and the associated connection should be freed. If communication for a particular connection is not completed in state 805, the connection is maintained for further communications between the requesting node such as client 110 and cache server 150 in state 810.

If it is determined that a connection shall be closed in state 805, the connection is closed in state 815 where the status 1010 of the connection tracking object 1005 is set inactive to indicate that the connection tracking object 1005 and corresponding tag are free for a new connection. In the preferred embodiment, a message associated with closing the connection tracking object 1005 is sent to the redirector 140 in state 820 so that the corresponding object in the redirector' s 140 connection tracking object array 1000 is also closed.

However, it should be noted that the cache server 150 and redirector 140 also maintain substantially identical connection tracking object arrays 1000 by each decoding the contents of the message to determine whether to open a new connection. For example, a connection and corresponding connection tracking object 1005 at the redirector 140 can be closed based on the detection of a FIN message, indicating that the message is last in line of a related stream of messages.

Fig. 9 is a flow chart illustrating how messages received from the cache server 150 are processed by the redirector 140. Messages are first received from the cache server 150 in state 905. It is then determined whether the message includes a FIN in state 925. If not, the message is routed to the network in state 935. If the message includes a FIN message in state 925, a "time wait" function is performed in state 927. Following time wait, the connection tracking object 1005 associated with the message is deleted in state 930 because the FIN message indicates the last of the data messages sent between a requesting node such as a client 110 and the cache server 150 for a particular connection. Based on this method of closing a connection in the redirector 20 and cache server 150, the associated connection tracking object arrays 1000 appropriately mirror each other.

Alternatively, a connection tracking object is closed based upon a direct order from the cache server 150. For example, if a connection tagging object is to be closed, the cache server optionally transmits a message to the redirector 140 to close a particular connection tracking object 1005.

Messages transmitted over the network in state 535 are "unmarked" with the tag before they are sent out over the network. For example, the tag is a code understood only by messages being passed between the cache server 150 and redirector 140. Hence, prior to transmission of the message over the network, the appropriate information from the connection tracking object is incorporated back into the message for proper routing if it is not already there. For example, the appropriate IP and TCP source and destination address are incorporated in the message. If a connection tag was appended to the network message, it is deleted so as not to interfere with subsequent routing of the message on the network.

As briefly described above, the implementation of a selective connectivity period provides a natural effect of controlling the new connection acceptance rate. For example, consider the case where the cache 150 is hunting for selective connections but the population of selective connections is low. In this case, the new connection SYNs allowed to be routed up to the cache server 150 are spaced at intervals of the selectivity period, t, plus the average SYN arrival interval .

Another important feature of the selectivity time period is that it provides a natural load control mechanism. For example, assume that the number of offered connections, (O_c) , is the actual number of connections passing through network 100. Also assume that the number of serviceable connections (S_) is the number of connections that the cache server 150 can actually service at any point in time. In general, the number of offered connections (O_c) will exceed the number of serviceable connections (S_c) since the cache server 150 has a finite capacity.

The goal is to obtain a higher hit rate for the cache server 150 as measured in the number of objects delivered from the cache server 150 as opposed to the number of objects which must be obtained from routes from the HTTP servers 19. Assuming that the number of offered connections exceeds the number of serviceable connections, setting the selectivity period to zero causes the cache server 150 will attempt to service all of the offered connections. On the other hand, if the selective connection period is set to a relatively high value, such as 100 milliseconds, the cache server 150 will likely service a connection count which is under its maximum capacity and thus spend most of its time hunting for SYNs that are on its selectivity list. Thus, by adjusting the selectivity period setting, one can provide an optimum connection load for the cache server 150.

To achieve this, the server 150 may preferably use a successive approximation approach by first setting the selectivity period to a predetermined value, such as fifty percent of a known maximum value, and then moving it up and down until the connection load runs just slightly below the maximum period. When this point is reached, the selectivity period is increased just enough to allow the server to run at an optimum rate . Turning attention to Fig. 12, there is shown an example of a plot of the number of serviced connections as a function of selectivity period. A horizontal dotted line 1200 indicates the maximum number of serviceable connections. By starting out a relatively high value, such as 100 milliseconds, for the selectivity period, the number of serviced connections is relatively low. However, as the selectivity period is reduced, the number of serviced connections gradually increases until a point is reached, such as at 1210, in which the maximum serviceable connections are reached. It is this setting or setting slightly below this which is the desired setting for the selectivity period. This will ensure that the cache server 150 is still attempting a sufficient number of new requests without becoming overloaded. This maximizes the hit rate in the cache. A natural time delay spacing for new connection requests is thus provided by setting the selectivity period to a value that slightly exceeds the system's maximum selectivity connection capacity.

However, it is also desirable to vary connection selectivity as a function of the load. In other words, it may be desirable to vary the selectivity period as the load increases or decreases. There are many advantages to this approach including:

(a) it is better to set the selectivity period to zero if the offered load fails to stress the server. This maximizes the cache utilization under light loads.

(b) for the moderate load case, it is desirable to set the Selectivity period at a lower period than the heavily loaded case. Again, the heavy load period may not be the optimum setting at moderate loads, (c) varying the selectivity period as a function of system state is also desirable. For example, if a system has been recently "cleared" and is re-converging (loading new objects into cache) it is likely that the optimum selectivity setting will be less. The conclusion that can be drawn is that the selectivity period is likely to be a non-linear value that will vary as a function of load. Moreover, this non-linear function will likely vary as a function of time. The first version of selectivity is likely to use the number of active or more desirable — offered connections to detect load. Thus, a typical function is suggested in Fig. 13.

It is expected that selectivity period setting as detected by the counter would therefore not be a constant but rather be a variable stored in a hardware memory that is loaded by either the server 150 or an intelligent subsystem, such as the NI (C) s 302. When a NIC 302 begins a new connection hunt phase, it uses the current connection count as an index into the selectivity array and reads out a period setting to use. In the above illustration suggests a possible non-linear selectivity array and reads out a period setting to use. In the example of Fig. 13, we assume that the server 150 runs efficiently at a maximum load of 100 to 1400 connections. as the load approaches maximum, the server 150 becomes increasingly selective by increasing the selectivity period. The selectivity period can actually reach infinity, meaning that the only requests to be processed are connections that have entries on the selectivity table 60.

To understand how various systems may require different optimized parameters, assume that the function, FI, shown in Fig. 13 represents one of many possible functions that can be loaded into the selectivity function table. Moreover, assume that the server has numerous functions that are loaded based on the system's "state." Examples of system state include the number of objects that are stored on the cache system, a rating of the current selectivity list (i.e., calculate the quality of the current selectivity list and apply the corresponding selectivity function for the given case) . Thus, we can apply various selectivity functions FI for various system "states" where the selectivity period is a function of the load, and the selectivity function shape is itself a function of systems state. Metrics which may be used for the system state include :

Ml — Number of objects cached

M2 — Average Hit rate (see below on feedback) M3 — Selective table rating - (how "good" is the selective table)

The selectivity function generator can be either a library of "functions" or derived using a simply generator function that linearizes the period function while using the system "state" for various coefficients.

It may also be desirable to vary the "Selectivity Function Curve" using a feedback system. This is based on the premise that the efficiency of the system may vary as a function of where the installation is. Thus, any two given cache systems may run more or less efficiently as a function of where they are installed. For this case it may be desirable for the cache to detect state metrics, and apply various functions until the optimum is found.

As a result of it being desirable to have different selectivity functions, the timer value is preferably set by a function running in the cache server 150. The cache server 150 also maintains a connection service processes which actually services active connections; that is, it accepts HTTP requests on active connections and provide the requested objects from the cache server 150 once active.

Another consideration is the actual generation of the contents of selectivity connection table (SCT) 430. The selective connection table generation process is preferably performed using an application that scans the cache object table stored in the cache and chooses IP addresses and sub-net masks for maximum efficiency. The algorithm used may consider both a number of stored objects weighted by a cache metric (e.g., popularity) . For example, one rating for selective IP address/sub-net address combinations can be provided as follows. Assume that a given selective address, K, (aa.bb.xx.xx of the form where aa.bb are the first two portions of the sub-net address and the xx.xx is the mask portion of the address since the last two octets are masked) . Also assume that N objects stored in the cache begin with the IP address aa.bb (or some other set described by the address K containing N stored objects) . The rating for a selective address K can therefore be defined as the sum of N object metrics R_AK = M₀₁+M₀₂+M₀₃+ ... +M_0N Hence, when the selective list generation algorithm rates a given address/sub-net combination, it does so based on the sum of the cache metrics (popularity) ratings for that combination.

There are other factors that can be included in this calculation. For example, the number of objects can be favored over the popularity rating as: R_AK = (M₀₁+M₀₂+M₀₃+...+M_0N)/N + cN Dividing the first terms by N yields the average metric for all of the objects contained in the set. This is added to N, times a constant, c, where c is adjusted to provide the proper weight for the number of objects relative to the average metric.

Finally, the sub-net mask can be considered where B equals the number of Is set in the mask:

R_AK = [(M₀₁+M₀₂+M₀₃+...+M_0N)/N + cN]/dB Dividing by the number of bits (times a constant, d) provides a lower rating for masks that are longer. This allows the most 'focused' sub-net combinations to yield better ratings.

It should be understood, however, that other techniques and algorithms may be used to determine the contents of the selective connection table 430.

While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims .

Claims

CLAIMSWhat is claimed is:

1. An apparatus comprising: a content server, for storing content objects and serving replicas of stored content objects in response to receiving a document request message; and a message redirector, associated with the content server, for selectively routing document request messages to the content server or to another server, the message redirector intercepting a document request message and examining server address contained therein, the message redirector further determining if the server address is contained on a local selective connection table contained in the message redirector, and selectively determining whether to route the intercepted request message to the content server or to another server.

2. An apparatus as in claim 1 wherein the content^" server additionally scans a list of content objects stored in the content server to identify a subset of the most requested content objects by server address, and then sends this list of server addresses to the message redirector to be stored on the local selective connection table.

An apparatus as in claim 1 wherein the content server is a cache server.

4. An apparatus as in claim 1 wherein the message redirector routes the intercepted request messages by examining one of a group of network layer addresses consisting of layer two (MAC) , layer three (TCP/IP), or layer four (port 80) addresses.

5. An apparatus as in claim 1 wherein the message redirector additionally comprises: selectivity process logic for, during a selectivity period, routing messages to the cache server only if the request message indicates an address associated with a content object in the selective connection table, and during a non-selective promiscuous period, routing any next request message to the cache server.

6. An apparatus as in claim 5 wherein the period of time during which the selective mode and non-selective mode are active may be adjusted.

7. An apparatus as in claim 5 wherein the period of time during which the selective mode and non-selective mode are active may be adjusted by setting the non-selective mode time period to an initially high value and then progressively shortening it until a point is reached near where a predetermined maximum possible number of serviceable connections are being handled by the content server.

An apparatus as in claim 1 wherein a server address associated with a content object is selected from the group consisting of a domain name, a full IP address or a sub-net mask address.

9. An apparatus as in claim 1 wherein the selective connection table has the effect of breaking a pool of available connections into a first group of selective connections derived from high probability hit rate objects and a second group of non-selective connections.

10. A method used in a network of server computers for optimizing the number of requested objects delivered from a cache server connected to the network, comprising the steps of: in the cache server, maintaining a list of demanded objects; in a message redirector, receiving periodic updates of a list of selective connections, and allowing only object requests to be routed up to the cache server during a selectivity time for requests that are on the list of selective connections, at the end of the selectivity period, allowing at least one new object request not on the list of selective connections to be routed up to the cache server; periodically generating a new selectivity list; and modifying a selectivity period until a point is reached near a maximum number of serviceable connections experienced in the cache server.

11. A method as in claim 10 wherein the cache server additionally performs the steps of: periodically scanning a list of content objects stored in the content server to identify a subset of the most requested objects by server address; and sending the list of server addresses to the message redirector to be used as the new selectivity list.

12. A method as in claim 10 wherein the message redirector additionally routes messages by performing the step of : routing object requests by examining one of a group of network layer addresses consisting of layer two (MAC) , layer three (TCP/IP) , or layer four (port 80) addresses.

13. A method as in claim 10 wherein the selectivity period may be adjusted.

14. A method as in claim 10 wherein the selectivity period may be adjusted by setting a non-selective mode time period to an initially high value and then progressively shortening it until a point is reached near where a predetermined maximum possible number of serviceable connections are being handled by the content server.

15. A method as in claim 10 wherein an object request is routed based upon a server address associated with a object as selected from the group consisting of a domain name, a full IP address or a sub-net mask address .

16. A method as in claim 10 wherein the selectivity functionality of the message redirector has the effect of breaking a pool of available connections into a first group of selective connections derived from high probability hit rate objects and a second group of non-selective connections.