Incremental high radix network scaling

Abstract

Efficient and highly-scalable network solutions are provided that utilize incremental scaling of switches, and devices connected to those switches, in an environment such as a data center. Embodiments may utilize multiple tiers of switches. Sets of switches in two different tiers may be initially connected to each other utilizing multiple connections. As network capacity needs within the computing environment increase, additional switches may be added to tiers. To connect the added switches to the switch network, the redundant connections may be utilized. Moving connections from one switch to another switch can free up ports to connect added switches in one of the tiers of switches to the switch network. The tiers of switches can be based on Clos networks, where the tiers of switches are fully connected, or other high radix or fat tree topologies that include oversubscription between tiers.

1 . (canceled) 
     
     
         2 . A method, comprising:
 providing a first set of switches, a second set of switches, a third set of switches, and a fourth set of switches in a data center;   connecting at least two ports of each switch of the first set of switches with at least two ports of each switch of the second set of switches;   connecting each switch of the first set of switches with each switch of the third set of switches;   receiving a request to adjust an arraignment of connections of the first set of switches, the second set of switches, the third set of switches, and the fourth set of switches   disconnecting, in a determined order, each switch of the second set of switches from numbered port locations with a corresponding switch and port of the first set of switches; and   connecting, with respect to the determined order, the disconnected switches with a respective switch of the fourth set of switches in an equivalent number port location.   
     
     
         3 . The method of  claim 2 , further comprising:
 providing a fifth set of switches; and   connecting each switch of the fifth set of switches with each switch of the second set of switches, the third set of switches, and the fourth set of switches.   
     
     
         4 . The method of  claim 3 , wherein connecting each switch of the fifth set of switches with each switch of the second set of switches, the third set of switches, and the fourth set of switches forms a fat tree network topology with or without oversubscription, wherein the fat tree network topology is a re-arrangeably non-blocking network topology. 
     
     
         5 . The method of  claim 3 , wherein connecting each switch of the fifth set of switches with ports in the second set of switches, the third set of switches, and the fourth set of switches, further includes:
 detecting an available port of a switch of one of the second set of switches, third set of switches, or fourth set of switches, wherein a port is available when the port goes from a connected state to a disconnected state; and   connecting to a respective port of a respective switch of the second set of switches, third set of switches, or fourth set of switches.   
     
     
         6 . The method of  claim 3 , wherein each of the switch of the first set of switches, the second set of switches, third set of switches, fourth set of switches, or the fifth set of switches comprises a commodity network switch or application-specific integrated circuit (ASIC). 
     
     
         7 . The method of  claim 3 , wherein a portion of the first set of switches are top of rack (TOR) switches connected to a plurality of host computing devices in a server rack, wherein a number of switches of the first set of switches equals a number of switches in the fifth set of switches, wherein a portion of the first set of switches are edge switches, a portion of the second set of switches are spine switches, a portion of the third set of switches are edge switches, a portion of the fourth set of switches are spine switches, and wherein a portion of the fifth set of switches are edge switches. 
     
     
         8 . The method of  claim 7 , wherein connecting each edge switch of the third set of switches with each spine switch of the second set of switches and each spine switch of the fourth set of switches results in a formation of a Clos network topology. 
     
     
         9 . The method of  claim 2 , wherein a portion of the first set of switches are top of rack (TOR) switches connected to a plurality of host computing devices in a server rack. 
     
     
         10 . The method of  claim 2 , wherein a portion of the first set of switches are top of rack (TOR) switches connected to a plurality of host computing devices in a server rack. 
     
     
         11 . A switch network in a data center, comprising:
 a first set of switches, each switch in the first set of switches including at least two ports;   a second set of, each switch in the second set of switches including at least two ports, wherein at least two ports of each switch of the first set of switches are connected with at least two ports of each switch of the second set of switches   a third set of switches connected with each switch of the first set switches utilizing one of the ports of the at least two ports between a respective switch of the first set of switches and a respective switch of the second set of switches, wherein one of the at least two connections is disconnected, in a determined order, from a port of the respective switch of the second set of switches and reconnected with a port of the respective switch of the third set of switches; and   a fourth set of switches connected, with respect to the determined order, with each disconnected switch of the second set of switches in an equivalent number port location.   
     
     
         12 . The switch network of  claim 11 , comprising:
 a fifth set of switches, wherein each switch of the fifth set of switches is connected with each switch of the second set of switches, the third set of switches, and the fourth set of switches.   
     
     
         13 . The switch network of  claim 12 , wherein connecting each switch of the fifth set of switches with each switch of the second set of switches, the third set of switches, and the fourth set of switches forms a fat tree network topology with or without oversubscription, wherein the fat tree network topology is a re-arrangeably non-blocking network topology. 
     
     
         14 . The switch network of  claim 12 , wherein a portion of the first set of switches are top of rack (TOR) switches connected to a plurality of host computing devices in a server rack, wherein a number of switches of the first set of switches equals a number of switches in the fifth set of switches, wherein a portion of the first set of switches are edge switches, a portion of the second set of switches are spine switches, a portion of the third set of switches are edge switches, a portion of the fourth set of switches are spine switches, and wherein a portion of the fifth set of switches are edge switches. 
     
     
         15 . The switch network of  claim 14 , wherein connecting each edge switch of the third set of switches with each spine switch of the second set of switches and each spine switch of the fourth set of switches results in a formation of a Clos network topology. 
     
     
         16 . The switch network of  claim 11 , wherein each switch of the first set of switches, the second set of switches, and the third set of switches comprises a commodity network switch or application-specific integrated circuit (ASIC). 
     
     
         17 . A computing system, comprising:
 a computing device processor;   a first set of switches, a second set of switches, a third set of switches, and a fourth set of switches in a data center; and   a memory device including instructions that, when executed by the computing device processor, cause the computing system to:
 connect at least two ports of each switch of the first set of switches with at least two ports of each switch of the second set of switches; 
 connect each switch of the first set of switches with each switch of the third set of switches; 
 receive a request to adjust an arraignment of connections of the first set of switches, the second set of switches, the third set of switches, and the fourth set of switches; 
 disconnect, in a determined order, each switch of the second set of switches from numbered port locations with a corresponding switch and port of the first set of switches; and 
 connect, with respect to the determined order, disconnected switches of the second set of switches with a respective switch of the fourth set of switches in an equivalent number port location. 
   
     
     
         18 . The computing system of  claim 17 , further including a fifth set of switches, wherein the instructions, when executed further enable the computing system to:
 connect each switch of the fifth set of switches with each switch of the second set of switches, the third set of switches, and the fourth set of switches.   
     
     
         19 . The computing system of  claim 18 , wherein connecting each switch of the fifth set of switches with each switch of the second set of switches, the third set of switches, and the fourth set of switches forms a fat tree network topology with or without oversubscription, wherein the fat tree network topology is a re-arrangeably non-blocking network topology. 
     
     
         20 . The computing system of  claim 18 , wherein the instructions, when executed to connect each switch of the fifth set of switches with ports in the second set of switches, the third set of switches, and the fourth set of switches, further enable the computing system to:
 detect an available port of a switch of one of the second set of switches, third set of switches, or fourth set of switches, wherein a port is available when the port goes from a connected state to a disconnected state; and   connect to a respective port of a respective switch of the second set of switches, third set of switches, or fourth set of switches.   
     
     
         21 . The computing system of  claim 18 , wherein each of the switch of the first set, the second set, third set, fourth set, or fifth set of switches comprises a commodity network switch or application-specific integrated circuit (ASIC).