Server load balancing system

• US 7,475,157 B1
• Filed: 09/13/2002
• Issued: 01/06/2009
• Est. Priority Date: 09/14/2001
• Status: Active Grant

• Current Assignee: Cloudblue LLC
• Original Assignee: Swsoft Holding Limited
• Inventors: Tormasov, Alexander
• Primary Examiner(s): Burgess; Glenton B
• Assistant Examiner(s): Meucci; Michael D
• Application Number:
  US10/242,810
• Time in Patent Office: 2,307 Days
• Field of Search: 709/227, 709/228, 709238-242, 714/807, 714/808
• US Class Current: 709/238
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/1014 based on the content of a request

Abstract

A system for maximizing the efficiency of a load balancing server for an asymmetric web farm utilizes a TCP stack and data packets to send and receive client service requests. An internal buffer enables each web server in the web farm to balance their loading based on the contents of the client service request. Data switching between the address portion of the user space and the address portion of the kernel space within the load balancing server is eliminated.

Claims

1. A load balancing server for balancing real servers in an asymmetric web farm, the load balancing server comprising:

a processor;

memory;

computer code loaded into the memory for performing functions of the load balancing server;

wherein the load balancing server;

receives a client request as a sequence of client TCP packets;

translates the sequence of client TCP packets to a server in a set of the real servers in the asymmetric web farm; and

translates a response of the real server as a sequence of server TCP packets back to the client,

wherein a service for the balancing and translating of the client TCP packets and the real server TCP packets is installed and running within a kernel space of the load balancing server,

wherein a real server from the set of real servers that processes the client request is chosen based on contents of the client TCP packets and based on network conditions,

wherein the load balancing server calculates a checksum for a header of the client TCP packet;

calculates a checksum for a header of the translated TCP packet having a new destination address;

generates a new checksum for the translated TCP packet, the translated TCP packet including a header and payload, the new checksum being generated by linear operations on the checksum for the client TCP packet, the checksum for the header of the client TCP packet and a checksum for the header of the translated TCP packet;

wherein the validity of the TCP packet, before and after translation, is preserved without additional verification; and

wherein each of the real servers services a specific class of the client requests based on file type of the client requests.

2. The load balancing server of claim 1, wherein said load balancing server uses functions of a TCP stack, including:

receiving an incoming connection, setting up an outgoing connection, closing a connection, receiving incoming data packets, and sending outgoing data packets.

3. The load balancing server of claim 2, wherein the load balancing server uses the function of the TCP stack to receive a client connection.

4. The load balancing server of claim 3, wherein the load balancing server receives all the incoming data packets of the client connection by using the functions of the TCP stack.

5. The load balancing server of claim 4, wherein the load balancing server continues storing the incoming data packets of the client connection in an internal queue until the real server designated to process the client request has been chosen based on contents of the incoming data packets.

6. The load balancing server of claim 5, wherein the load balancing server determines the real server to respond to the client request according to preset operating rules.

7. The load balancing server of claim 6, wherein the preset operating rules include:

contents of the client service request of the internal queue;

an IP address of a client determined by titles of packets of the internal queue;

loading level of processors of web servers assigned for client service request processing; and

size of memory of the real servers assigned for client request processing.

8. The load balancing server of claim 1, wherein the load balancing server creates a TCP connection to the real server assigned for client service request processing by using functions of a TCP stack, the functions comprising:

receiving an incoming connection, setting up an outgoing connection,

closing a connection,

receiving incoming data packets, and

sending outgoing data packets.

9. The load balancing server of claim 1, wherein said load balancing server operates a TCP checksum packet translating algorithm for packet dispatching, wherein said TCP checksum packet translating algorithm formulates a TCP packet data checksum through a checksum of a TCP packet header.

10. The load balancing server of claim 9, wherein, to provide consistency, the load balancing server updates TCP headers of data packets and sends the data packets stored in an internal queue to the real server assigned for client request processing using the TCP checksum packet translating algorithm and functions of a TCP stack.

11. The load balancing server of claim 10, wherein the load balancing server terminates other connections to the real servers by invoking functions of the TCP stack when the connections between the client and the load balancing server and between the load balancing server and the real server chosen to respond to the client requests are closed.

12. The load balancing server of claim 9, wherein, to provide consistency, the load balancing server, when the connections from the client to the load balancing server and from the load balancing server to the real server assigned for client service request processing are open, updates TCP headers of data packets and relocates the TCP data packets received by one of the connections to the other outgoing connection using the TCP checksum translating algorithm and functions of a TCP stack.

13. The load balancing server of claim 1, wherein the load balancing server, upon receiving the client sequence, places the client sequence in a memory buffer.

14. The load balancing server of claim 13, wherein the load balancing server, upon placing the sequence in the memory buffer, parses the client TCP packets to analyze client requests, the parsing being installed and running in kernel space.

15. The load balancing server of claim 13, wherein the load balancing server, upon parsing the client TCP packets, modifies headers of the client TCP packets to ensure consistency of the client TCP packets.

16. The load balancing server of claim 1, wherein superusers that receive priority service from the real servers are identified based on their IP address.

17. A computer storage medium having computer executable program logic stored thereon, the computer executable program logic executing on a processor for balancing load of real servers in a web farm, the computer program logic comprising:

(a) computer program code means for receiving a client request as a sequence of client TCP packets;

(b) computer program code means for translating the sequence of the client TCP packets to a server in a set of the real servers in the web farm; and

(c) computer program code means for translating a response of the real server as a sequence of the real server TCP packets back to the client,

wherein a real server from the set of real servers that processes the client request is chosen based on contents of the client TCP packets and based on network conditions,

wherein the load balancing server calculates a checksum for a header of the client TCP packet;

calculates a checksum for a header of the translated TCP packet having a new destination address;

generates a new checksum for the translated TCP packet, the translated TCP packet including a header and payload, the new checksum being generated by linear operations on the checksum for the client TCP packet, the checksum for the header of the client TCP packet and a checksum for the header of the translated TCP packet;

wherein the validity of the TCP packet, before and after translation, is preserved without additional verification; and

wherein each of the real servers services a specific class of the client requests based on file type of the client requests.

18. The computer storage medium of claim 17, wherein the web farm is an asymmetric web farm, and wherein the computer program code means (a), (b) and (c) are installed and running within a kernel space of a load balancing server.

19. A method for balancing load of real servers in an asymmetric web farm, comprising:

receiving a client request as a sequence of client TCP packets;

translating the sequence of client TCP packets to a server in a set of the real servers in the asymmetric web farm; and

translating a response of the real server as a sequence of the real server TCP packets back to the client,

wherein a service for balancing and translating of the client TCP packets and the real server TCP packets is installed and running within a kernel space of a load balancing server,

wherein a real server from the set of real servers that processes the client request is chosen based on contents of the client TCP packets and based on network conditions;

calculating a checksum for a header of the client TCP packet;

calculating a checksum for a header of the translated TCP packet having a new destination address;

generating a new checksum for the translated TCP packet, the translated TCP packet including a header and payload, the new checksum being generated by linear operations on the checksum for the client TCP packet, the checksum for the header of the client TCP packet and a checksum for the header of the translated TCP packet;

wherein the validity of the TCP packet, before and after translation, is preserved without additional verification; and

wherein each of the real servers services a specific class of the client requests based on file type of the client requests.

20. A load balancing server for balancing real servers in an asymmetric web farm, the load balancing server comprising:

a processor;

memory;

computer code loaded into the memory for performing functions of the load balancing server;

a load balancing module; and

a TCP packet translating module that receives a client request as a sequence of client TCP packets, translates the sequence of the client TCP packets to a real server of the asymmetric web farm, and translates a response of the real server as a sequence of the real server TCP packets back to the client,

wherein a real server is chosen from a set of real servers of the asymmetric web farm based on network conditions,

wherein the load balancing server calculates a checksum for a header of the client TCP packet;

calculates a checksum for a header of the translated TCP packet having a new destination address;

generates a new checksum for the translated TCP packet, the translated TCP packet including a header and payload, the new checksum being generated by linear operations on the checksum for the client TCP packet, the checksum for the header of the client TCP packet and a checksum for the header of the translated TCP packet;

wherein the validity of the TCP packet, before and after translation, is preserved without additional verification.

21. The load balancing server of claim 20, wherein:

the load balancing module is installed and running within a kernel space of an operating system of the load balancing server; and

wherein the TCP packet translating module is installed and running within a kernel space of the load balancing server,

wherein the real server is chosen from the set of real servers based on file type of the client requests.

22. A load balancing server for balancing real servers in an asymmetric web farm, wherein the load balancing server comprises:

a processor;

memory;

computer code loaded into the memory for performing functions of the load balancing server;

wherein the load balancing server;

receives a client response as a sequence of client TCP packets;

translates the sequence of client TCP packets to a server in a set of the real servers in the asymmetric web farm; and

translates a response of the real server as a sequence of server TCP packets back to the client,

wherein a service for the balancing and translating of the client TCP packets and the real server TCP packets is installed and running within a kernel space of the load balancing server, and

wherein some clients are assigned higher priority for responding to their client requests based on client IP addresses,

wherein servers that respond to the client requests are chosen based on network conditions, and

wherein the load balancing server calculates a checksum for a header of the client TCP packet;

calculates a checksum for a header of the translated TCP packet having a new destination address;

generates a new checksum for the translated TCP packet, the translated TCP packet including a header and payload, the new checksum being generated by linear operations on the checksum for the client TCP packet, the checksum for the headed of the client TCP packet and a checksum for the header of the translated TCP packet;

wherein the validity of the TCP packet, before and after translation, is preserved without additional verification.

23. A load balancing server for balancing real servers in a farm of server applications, wherein the load balancing server comprises:

a processor;

memory;

computer code loaded into the memory for performing functions of the load balancing server, wherein the load balancing server;

receives a client request as a sequence of client TCP packets;

translates the sequence of client TCP packets to the real server applications in the farm; and

translates a response of the real server applications as a sequence of TCP packets back to the client,

wherien the load balancing server calculates a checksum for a header of the client TCP packet;

calculates a checksum for a header of the translated TCP packet having a new destination address;

generates a new checksum for the translated TCP packet, the translated TCP packet including a header and payload, the new checksum being generated by linear operations on the checksum for the client TCP packet, the checksum for the header of the client TCP packet and a checksum for the header of the translated TCP packet;

wherein the validity of the TCP packet, before and after translation, is preserved without additional verification;

wherein a real server application from the set of real server applications that processes the client request is chosen based on requested service in message contents of the client request,

wherein each of the real server applications services a specific class of the client requests that is selected based on the message contents, and generates responses to the incoming client requests.

24. The load balancing server of claim 23, wherein:

the farm is an asymmetric farm,

wherien each of the real server applications services a specific class of the client requests that is selected based and based on network conditions, and

wherein superusers that receive priority service from the real servers are identified based in their IP address.