DYNAMIC SCALING OF CLOUD GATEWAYS CONNECTING CLOUD MACHINES TO ON-PREMISES MACHINES BASED ON STATISTICS OF THE CLOUD GATEWAYS

1 . A method for deploying cloud gateways between a set of cloud machines in a first network and a set of on-premises machines in an external network, the method comprising:

collecting a set of statistics for a first cloud gateway used to connect the set of cloud machines and the set of on-premises machines;

analyzing the set of statistics to determine that a second cloud gateway is needed to connect the set of cloud machines and the set of on-premises machines;

identifying a subset of the set of cloud machines; and

distributing a set of one or more forwarding rules to the subset of cloud machines to forward a set of data message flows from the subset of cloud machines to the set of on-premises machines through the second cloud gateway.

2 . The method of claim 1 further comprising, before distributing the set of forwarding rules, deploying the second cloud gateway in the first network.

3 . The method of claim 1 , wherein the set of data message flows is a first set of data message flows, and cloud machines in the set of cloud machines not in the identified subset of cloud machines continue to forward a second set of data message flows to the set of on-premises machines through the first cloud gateway.

4 . The method of claim 1 , wherein the set of statistics comprises statistics specifying bandwidth use of the first cloud gateway by the set of cloud machines at different times.

5 . The method of claim 4 , wherein the statistics specify bandwidth use of a particular uplink interface of the first cloud gateway by the set of cloud machines to connect to the set of on-premises machines.

6 . The method of claim 4 , wherein analyzing the set of statistics comprises determining that a predicted future bandwidth use of the first cloud gateway by the set of cloud machines exceeds a particular threshold.

7 . The method of claim 4 , wherein the set of statistics is a first set of statistics, the method further comprising collecting a second set of statistics for the set of cloud machines, wherein analyzing the first set of statistics comprises analyzing the first and second set of statistics to determine that the second cloud gateway is needed.

8 . The method of claim 7 , wherein identifying the subset of cloud machines comprises analyzing the second set of statistics to determine that the subset of cloud machines consumes more bandwidth of the first cloud gateway than other cloud machines in the set of cloud machines.

9 . The method of claim 1 , wherein collecting the set of statistics comprises retrieving the set of statistics from a data store.

10 . The method of claim 9 further comprising:

before retrieving the set of statistics, iteratively collecting subsets of statistics for the first cloud gateway at specified time intervals; and

storing the collected subsets of statistics in the data store, wherein the set of statistics comprises different subsets of statistics.

11 . The method of claim 9 , wherein the data store is a time-series database.

12 . The method of claim 1 , wherein the first network is a cloud datacenter and the external network is an on-premises datacenter.

13 . The method of claim 12 , wherein the cloud datacenter and the on-premises are respectively first and second software defined datacenters (SDDCs).

14 . The method of claim 1 , wherein the cloud machines are virtual machines (VMs) operating on host computers in the first network.

15 . A non-transitory machine readable medium storing a program for execution by at least one processing unit for deploying cloud gateways between a set of cloud machines in a first network and a set of on-premises machines in an external network, the program comprising sets of instructions for:

collecting a set of statistics for a first cloud gateway used to connect the set of cloud machines and the set of on-premises machines;

analyzing the set of statistics to determine that a second cloud gateway is needed to connect the set of cloud machines and the set of on-premises machines;

identifying a subset of the set of cloud machines; and

distributing a set of one or more forwarding rules to the subset of cloud machines to forward a set of data message flows from the subset of cloud machines to the set of on-premises machines through the second cloud gateway.

16 . The non-transitory machine readable medium of claim 15 , wherein the program further comprises sets of instructions for, before distributing the set of forwarding rules, deploying the second cloud gateway in the first network.

17 . The non-transitory machine readable medium of claim 15 , wherein the set of data message flows is a first set of data message flows, and cloud machines in the set of cloud machines not in the identified subset of cloud machines continue to forward a second set of data message flows to the set of on-premises machines through the first cloud gateway.

18 . The non-transitory machine readable medium of claim 15 , wherein the set of statistics comprises statistics specifying bandwidth use of the first cloud gateway by the set of cloud machines at different times.

19 . The non-transitory machine readable medium of claim 18 , wherein the set of instructions for analyzing the set of statistics comprises sets of instructions for determining that a predicted future bandwidth use of the first cloud gateway by the set of cloud machines exceeds a particular threshold.

20 . The non-transitory machine readable medium of claim 18 , wherein:

the set of statistics is a first set of statistics,

the program further comprises sets of instructions for collecting a second set of statistics for the set of cloud machines, and

the set of instructions for analyzing the first set of statistics comprises sets of instructions for analyzing the first and second set of statistics to determine that the second cloud gateway is needed.