Tearing down virtual machines implementing parallel operators in a streaming application based on performance

• US 10,078,530 B2
• Filed: 01/13/2017
• Issued: 09/18/2018
• Est. Priority Date: 10/10/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Bragstad, Lance, Branson, Michael J., Cao, Bin, Carey, James E., Odden, Mathew R.
• Primary Examiner(s): Puente, Emerson
• Assistant Examiner(s): Do, Steven
• Application Number:
  US15/406,587
• Publication Number:
  US 20170147386A1
• Time in Patent Office: 613 Days
• Field of Search: None
• CPC Class Codes: G06F 2009/45562 Creating, deleting, cloning virtual machine instances
• CPC Class Codes: G06F 2009/45575 Starting, stopping, suspending or resuming virtual machine instances
• CPC Class Codes: G06F 2009/45591 Monitoring or debugging support
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: H05K 999/99 dummy group

Abstract

A streams manager monitors performance of parallel portions of a streaming application implemented in multiple virtual machines (VMs). When the performance provided by the multiple VMs is no longer needed, one or more of the VMs can be torn down. The performance of the VMs is monitored. When the least performing VM can be torn down, it is torn down. When the least performing VM cannot be torn down, information regarding a better performing VM is gathered, and it is determined whether the least performing VM can be made more similar to the better performing VM. When the least performing VM can be made more similar to the better performing VM, the least performing VM is changed to improve its performance, and the better performing VM is torn down.

Claims

1. An apparatus comprising:

at least one processor;

a memory coupled to the at least one processor;

a streaming application residing in the memory and executed by the at least one processor, the streaming application comprising a flow graph that includes a plurality of operators that process a plurality of data tuples, wherein the flow graph comprises a first set of operators implemented in a first virtual machine and a second set of operators implemented in a second virtual machine, wherein the first set of operators and the second set of operators include similar operators and connections between operators and operate on data tuples from a single data stream in parallel and produce data tuples that are sent to a same operator in the streaming application; and

a streams manager residing in the memory and executed by the at least one processor, the streams manager monitoring performance of the streaming application, and when performance of the streaming application indicates processing data tuples in parallel by the first set of operators and the second set of operators is no longer needed, the streams manager determines which of the first virtual machine and second virtual machine is least performing, and when the least performing virtual machine can be torn down, the streams manager tears down the least performing virtual machine, wherein when the least performing virtual machine cannot be torn down, the streams manager gathers data regarding a better performing virtual machine operating on the flow graph in parallel with the least performing virtual machine, and when the least performing virtual machine can be made more similar to the better performing virtual machine in its current location, the streams manager makes at least one change to the least performing virtual machine to improve performance of the least performing virtual machine and tears down the better performing virtual machine, wherein when the least performing virtual machine cannot be made more similar to the better performing virtual machine in its current location, the streams manager determines when the least performing virtual machine can be migrated to a location where the least performing virtual machine can be made more similar to the better performing virtual machine, migrates the least performing virtual machine to the location, makes at least one change to the least performing virtual machine to improve performance of the least performing virtual machine, and tears down the better performing virtual machine.

2. The apparatus of claim 1 wherein when the least performing virtual machine cannot be migrated, the streams manager tears down the better performing virtual machine.

3. The apparatus of claim 1 wherein the streams manager logs performance data for each virtual machine that includes operators in the flow graph.

4. The apparatus of claim 3 wherein the streams manager uses the logged performance data to determine a desired configuration for a new virtual machine that implements a new set of operators that operate on data tuples in parallel with operators in at least one existing virtual machine.

5. The apparatus of claim 1 wherein the streams manager tears down the least performing virtual machine by sending a request to a cloud manager to tear down the least performing virtual machine.

6. The apparatus of claim 1 wherein the streams manager tears down the better performing virtual machine by sending a request to a cloud manager to tear down the better performing virtual machine.

8. The method of claim 1 wherein when the least performing virtual machine cannot be migrated, tearing down the better performing virtual machine.

9. The method of claim 1 further comprising logging performance data for each virtual machine that includes operators in the flow graph.

10. The method of claim 9 further comprising using the logged performance data to determine a desired configuration for a new virtual machine that implements a new set of operators that operate on data tuples in parallel with operators in at least one existing virtual machine.

11. The method of claim 1 wherein tearing down the least performing virtual machine comprises sending a request to a cloud manager to tear down the least performing virtual machine.

12. The method of claim 1 wherein tearing down the better performing virtual machine comprises sending a request to a cloud manager to tear down the better performing virtual machine.

7. A computer-implemented method executed by at least one processor for managing a streaming application, the method comprising:

executing a streaming application that comprises a flow graph that includes a plurality of operators that process a plurality of data tuples, wherein the flow graph comprises a first set of operators implemented in a first virtual machine and a second set of operators implemented in a second virtual machine, wherein the first set of operators and the second set of operators include similar operators and connections between operators and operate on data tuples from a single data stream in parallel and produce data tuples that are sent to a same operator in the streaming application;

monitoring performance of the streaming application;

when performance of the streaming application indicates processing data tuples in parallel by the first set of operators and the second set of operators is no longer needed, determining which of the first virtual machine and second virtual machine is least performing;

determining that the least performing virtual machine cannot be torn down;

gathering data regarding a better performing virtual machine operating on the flow graph in parallel with the least performing virtual machine;

determining that the least performing virtual machine cannot be made more similar to the better performing virtual machine in its current location;

determining when the least performing virtual machine can be migrated to a location where the least performing virtual machine can be made more similar to the better performing virtual machine;

migrating the least performing virtual machine to the location;

making at least one change to the least performing virtual machine to improve performance of the least performing virtual machine; and

tearing down the better performing virtual machine.

13. An apparatus comprising:

at least one processor;

a memory coupled to the at least one processor;

a streaming application residing in the memory and executed by the at least one processor, the streaming application comprising a flow graph that includes a plurality of operators that process a plurality of data tuples, wherein the flow graph comprises a first set of operators implemented in a first virtual machine and a second set of operators implemented in a second virtual machine, wherein the first set of operators and the second set of operators include similar operators and connections between operators and operate on data tuples in parallel; and

a streams manager residing in the memory and executed by the at least one processor, the streams manager monitoring performance of the streaming application, and when performance of the streaming application indicates processing data tuples in parallel by the first set of operators and the second set of operators is no longer needed, the streams manager determines which of the first virtual machine and second virtual machine is least performing, and when the least performing virtual machine can be torn down, the streams manager tears down the least performing virtual machine, wherein when the least performing virtual machine cannot be torn down, the streams manager gathers data regarding a better performing virtual machine operating on the flow graph in parallel with the least performing virtual machine, and when the least performing virtual machine can be made more similar to the better performing virtual machine in its current location, the streams manager makes at least one change to the least performing virtual machine to improve performance of the least performing virtual machine and tears down the better performing virtual machine, wherein when the least performing virtual machine cannot be made more similar to the better performing virtual machine in its current location, the streams manager determines when the least performing virtual machine can be migrated to a location where the least performing virtual machine can be made more similar to the better performing virtual machine, migrates the least performing virtual machine to the location, makes at least one change to the least performing virtual machine to improve performance of the least performing virtual machine, and tears down the better performing virtual machine.