Reducing power consumption in a fat-tree network

1. A method for communication, comprising:

configuring a multi-level fat-tree network to comprise at least three levels of switches, including multiple modules arranged externally in a tree topology, each module containing a respective group of the switches arranged in an internal tree extending over at least two of the levels of the network;

selecting, by a network manager, a subset of the modules to be active in carrying the communication traffic; and

operating the network so as to convey communication traffic among the switches via the active modules, while the modules that are not in the selected subset remain inactive.

2. The method according to claim 1 , and comprising selecting within at least one of the active modules a sub-group of the switches to be active, while the switches that are not in the selected sub-group within the at least one of the active modules remain inactive.

3. The method according to claim 1 , wherein the multi-level fat-tree network comprises at least first, second, third and fourth levels of the switches, and wherein the multiple modules comprises lower-level modules comprising respective lower-level groups of the switches in at least the first and second levels and upper-level modules comprising respective upper-level groups of the switches in at least the third and fourth levels.

4. The method according to claim 1 , wherein selecting the subset comprises:

estimating a characteristic of communication traffic to be carried by the fat-tree network;

responsively to the estimated characteristic, selecting, according to a predetermined selection order, certain of the spine switches in a highest level of the network to be active in carrying the communication traffic; and

in each of the levels of the spine switches below the highest level, selecting the spine switches to be active based on the selected spine switches in a next-higher level.

5. The method according to claim 4 , wherein the selection order is determined so as to maximize a number of the spine switches in intermediate levels of the network that can be inactivated while still maintaining full connectivity between leaf switches that are interconnected by the spine switches in the network.

6. The method according to claim 1 , and comprising receiving computing jobs that are to be executed by a plurality of computing nodes connected to leaf switches of the network, wherein the leaf switches are interconnected by links with the spine switches,

wherein selecting the subset comprises scheduling the jobs to run on respective groups of the computing nodes, wherein the groups are selected so as to enable one or more of the spine switches to be deactivated during execution of at least one of the jobs.

7. Apparatus for communication, comprising:

a plurality of switches arranged in a multi-level fat-tree network comprising at least three levels of the switches, including multiple modules arranged externally in a tree topology, each module containing a respective group of the switches arranged in an internal tree extending over at least two of the levels of the network; and

a network manager, which is configured to select a subset of the modules to be active in carrying the communication traffic, so as to cause the network to convey communication traffic among the switches via the active modules, while the modules that are not in the selected subset remain inactive.

8. The apparatus according to claim 7 , wherein the network manager is configured to select a sub-group of the switches within at least one of the active modules to be active, while the switches that are not in the selected sub-group within the at least one of the active modules remain inactive.

9. The apparatus according to claim 7 , wherein the multi-level fat-tree network comprises at least first, second, third and fourth levels of the switches, and wherein the multiple modules comprises lower-level modules comprising respective lower-level groups of the switches in at least the first and second levels and upper-level modules comprising respective upper-level groups of the switches in at least the third and fourth levels.

10. The apparatus according to claim 7 , wherein the network manager is configured to select, according to a predetermined selection order and responsively to the estimated characteristic of communication traffic to be carried by the network, certain of the spine switches in a highest level of the network to be active in carrying the communication traffic, and in each of the levels of the spine switches below the highest level, to select the spine switches to be active based on the selected spine switches in a next-higher level.

11. The apparatus according to claim 10 , wherein the selection order is determined so as to maximize a number of the spine switches in intermediate levels of the network that can be inactivated while still maintaining full connectivity between leaf switches that are interconnected by the spine switches in the network.

12. The apparatus according to claim 7 , wherein the network manager is configured to receive computing jobs that are to be executed by a plurality of computing nodes connected to leaf switches of the network, wherein the leaf switches are interconnected by links with the spine switches, and to schedule the jobs to run on respective groups of the computing nodes,

wherein the groups are selected so as to enable one or more of the spine switches to be deactivated during execution of at least one of the jobs, and to cause the jobs to be executed on the respective groups of the computing nodes as scheduled while deactivating the one or more of the spine switches.

13. The method according to claim 1 , wherein the modules that are inactive are powered down.

14. The method according to claim 13 , wherein the modules that are inactive are in a low-power state.

15. The method according to claim 13 , wherein the modules that are inactive are powered off.

16. The apparatus according to claim 7 , wherein the modules that are inactive are powered down.

17. The apparatus according to claim 16 , wherein the modules that are inactive are in a low-power state.

18. The apparatus according to claim 16 , wherein the modules that are inactive are powered off.