Electrical computers and digital processing systems: multicomput – Computer network managing – Network resource allocating
Reexamination Certificate
1998-06-23
2001-01-30
Rinehart, Mark H. (Department: 2756)
Electrical computers and digital processing systems: multicomput
Computer network managing
Network resource allocating
C709S241000, C709S201000
Reexamination Certificate
active
06182139
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to network servers, and more particularly to URL-based server-assignment on the client-side.
BACKGROUND OF THE INVENTION
Many Internet users can testify to the utter frustration when a “SERVER NOT RESPONDING” error message is displayed on their browser while trying to connect to a web site. Users often blame the company that administers the unavailable web site, despite large investments in replicated servers. Internet technologies are increasingly used for mission-critical Intranet applications that require high levels of reliability. An intelligently-designed web-site architecture with better fault-tolerance is needed.
The parent application described a server farm that uses a single, virtual IP address for all server machines. The client first connects to a load-balancer at the server farm that receives all incoming packets to the virtual IP address. The load-balancer makes the initial connection with the client, saving the packets used to make the connection. Once the connection is made and acknowledged, the client sends the data request in a packet that contains a Universal Resource Locator, or URL.
The load-balancer extracts the URL from the packet. The URL specifies a file or service at the web-server farm. The load-balancer then selects only those servers that have the resource requested in the URL. Load-balancing is then performed among these servers.
Since load-balancing waits until the URL is received, load-balancing depends on the resource requested, not merely on the load of each server. The server farm can be heterogeneous—the entire content of the web site does not have to be mirrored to each server's hard disk. Some files and resources may be located on a single node or a few nodes. Other web sites resources may include dedicated servers with specific resources such as databases or built-in application-programming interfaces (API's) to interface with user-defined programs, or software licenses to run particular programs on particular servers. Other servers may support the SMTP or POP3 post-office-protocols for e-mail, the NNTP protocol for newsgroups, etc. These specialized resources can reside on just a few of the servers.
TCP State Migration
The connection is transferred from the load-balancer to the assigned server once the URL is extracted. In this process, called TCP state migration, the stored packets from the client are replayed by the load-balancer to the assigned server. The acknowledgement packets from the assigned server are captured by the load-balancer and deleted so that the client is unaware that the connection has been transferred from the load-balancer to the assigned server. Future packets from the client are first routed to the load-balancer, and then passed through to the assigned server. The assigned server uses the virtual IP address instead of its own IP address as the source address for packets returned to the client.
TCP state migration is described in much more detail in the parent, U.S. Pat. No. 5,774,660, hereby incorporated by reference.
Fault-Tolerant Web Site—
FIG. 1
The parent application described a fault-tolerant web farm using a back-up load balancer.
FIG. 1
is a diagram of a fault-tolerant web site with a back-up load balancer and dual Internet connections. Browser client
10
sends requests through Internet
66
with a virtual IP address for the whole web site. Incoming packets with the virtual IP address are routed to load balancer
70
over local LAN
144
. Local LAN
144
may contain routers, switches, and hubs (not shown) when servers are located on separate network nodes. Most large server farms have very complex interconnections within the local LAN network. Local LAN
144
connects to Internet
66
through Internet connection
142
which directly connects to Internet connection router
140
, and through Internet connection
148
, which is connected to Internet connection router
146
.
Connections
142
,
148
provide connection to Internet
66
. A primary load balancer
70
is used to direct and load balance connections across servers
51
,
52
,
55
,
56
. A backup load balancer
70
′ is also provided to take over operation should primary load balancer
70
fail. These load balancers are located on separate servers to lessen the chance that both fail at the same time. Backup load balancer
70
′ closely monitors primary load balancer
70
to detect a failure.
It is somewhat undesirable to have a back-up load balancer. If the primary and back-up load-balancers crash or otherwise becomes unavailable, client packets are no longer forwarded to their servers, causing the clients to hang. A technique to avoid this client hang caused by a load-balancer crash is desired.
Incoming Packets Through Load-Balancer
FIG. 1
shows that incoming packets from client
10
are all routed through load-balancer
70
, since a virtual IP address is used as the destination address. Load-balancer
70
changes the address of these incoming packets to direct them to the assigned server, such as server
52
. Server
52
then sends the data back to client
10
, using the virtual IP address as the source.
Although server
52
sends outgoing packets directly to client
10
, the incoming packets still all go through load-balancer
70
. This still creates somewhat of a network bottleneck, although certainly not as severe as the prior art since only the smaller incoming packets are routed through load-balancer
70
. It is desirable to remove this bottleneck.
FIG. 2
shows a client attempting to connect to a failed server. Similar reference numbers are used as described for FIG.
1
. Client
10
attempts to open a connection to load-balancer
70
, which is migrated over to server
52
. Packets from client
10
are not responded to by server
52
since server
52
has crashed. Client
10
receives no reply from server
52
, so client
10
eventually displays an error message to the user such as “Server Not Responding”. Otherwise, the client times out while waiting for a response, and displays the error message “Server Down or Unreachable”, or “Connection Timed Out”.
Another common error is that the client uses a stale reference—a URL to a web page that no longer exists or has been relocated. It is desirable to have URL automatically translated to the relocated web page to avoid this error.
It is desired to reduce the frequency of “SERVER NOT RESPONDING” and “SERVER TIMED OUT” messages that Internet users often receive. A more efficient and fault-tolerant web-site architecture that avoids the data bottleneck and single point of failure at the load-balancer at the web site is desired. It is desired for the client to perform load-balancing or server assignment and for the TCP connection to be migrated from the client to the assigned server. Client-side server-assignment is desirable.
WAN load balancing is also desirable. To minimize client latency the client should be served by the best responding server. Responding is a function of both server load and network latency. Minimizing latency is desirable since minimal-latency paths tend to go around Internet bottlenecks, because server routes though bottlenecks are slower. This improves the overall performance of the Internet and/or company WAN links.
SUMMARY OF THE INVENTION
A client-side dispatcher migrates a connection for a high-level client application from an original server to a relocated server. The client-side dispatcher has a session table for storing an original server address of the original server and a relocated server address of the relocated server. An input receives first packets from the high-level client application. The first packets have the original server address as a destination address. An output outputs to a network second packets. The second packets have the relocated server address as the destination address.
A local packet storage is for storing the first packets. An address translator converts the destination address of the first packets from the original server address to the relocated server address to gener
Auvinen Stuart T.
Jaroenchonwanit Bunjob
Resonate Inc.
Rinehart Mark H.
LandOfFree
Client-side resource-based load-balancing with... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Client-side resource-based load-balancing with..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Client-side resource-based load-balancing with... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2478446