Distributed anomaly detection using combinable measurement value summaries

1. A method comprising:

sending, by a computing device to a central computing device as the computing device initiates, a request for a current global combined quantile sketch data structure comprising a plurality of local quantile sketch data structures received by the central computing device from a plurality of local computing devices;

obtaining, by the computing device over a first period of time, a first plurality of measurement values of a first item being measured, wherein the first item corresponds to a physical machine comprising a rotating shaft, and the first plurality of measurement values correspond to rotation speeds of the rotating shaft;

adding, over the first period of time by the computing device, each of the first plurality of measurement values to a first local quantile sketch data structure, the first local quantile sketch data structure having a bounded physical size and being configured to maintain a summarization of the first plurality of measurement values;

sending the first local quantile sketch data structure to the central computing device;

receiving, by the computing device from the central computing device, a second global combined quantile sketch data structure comprising the first local quantile sketch data structure and one or more other local quantile sketch data structures;

obtaining, by the computing device, a plurality of new measurement values of the first item;

for each new measurement value of the plurality of new measurement values:

adding, by the computing device, the new measurement value to a second local quantile sketch data structure;

adding, by the computing device, the new measurement value to the second global combined quantile sketch data structure; and

querying, by the computing device, the second global combined quantile sketch data structure to determine whether the new measurement value is an anomalous measurement value; and

reconfiguring, responsive to determining that at least one new measurement value is an anomalous measurement value, an operating behavior of the physical machine.

2. The method of claim 1 further comprising:

obtaining, by the computing device over a second period of time, a second plurality of measurement values of the first item being measured;

adding, by the computing device, each of the second plurality of measurement values to a second local quantile sketch data structure, the second local quantile sketch data structure configured to maintain a summarization of the second plurality of measurement values;

obtaining, by the computing device, a second new measurement value of the first item; and

querying a combination of the second global combined quantile sketch data structure and the second local quantile sketch data structure to determine whether the second new measurement value of the first item is an anomalous measurement value.

3. The method of claim 2 further comprising:

for each respective measurement value of the second plurality of measurement values:

adding, to the second global combined quantile sketch data structure, the respective measurement value; and

adding, to the second local quantile sketch data structure, the respective measurement value.

4. The method of claim 2 further comprising:

for each respective measurement value of the second plurality of measurement values:

combining the second global combined quantile sketch data structure with the second local quantile sketch data structure to generate a new combined measurement value summary; and

querying the new combined measurement value summary to determine whether the respective measurement value is an anomalous measurement value.

5. The method of claim 1 wherein the first local quantile sketch data structure comprises a t-digest, and the current global combined quantile sketch data structure comprises a t-digest.

6. The method of claim 1 wherein the first local quantile sketch data structure remains a fixed size over the first period of time irrespective of a number of measurements.

7. The method of claim 1 further comprising:

receiving, by the computing device from the central computing device, a request to provide the first local quantile sketch data structure to the central computing device; and

in response to the request, sending the first local quantile sketch data structure to the central computing device.

8. The method of claim 1 wherein the first plurality of measurement values is obtained periodically over the first period of time.

9. The method of claim 1 wherein the first item being measured comprises a response time of a computing environment resource, and wherein obtaining the first plurality of measurement values of the first item being measured comprises determining a first plurality of response times associated with the computing environment resource.

10. The method of claim 9 wherein the computing environment resource comprises one of an application and a database.

11. A non-transitory computer-readable storage medium that includes executable instructions to cause a computing device to:

send, to a central computing device as the computing device initiates, a request for a current global combined quantile sketch data structure comprising a plurality of local quantile sketch data structures received by the central computing device from a plurality of local computing devices;

obtain, over a first period of time, a first plurality of measurement values of a first item being measured, wherein the first item corresponds to a physical machine comprising a rotating shaft, and the first plurality of measurement values correspond to rotation speeds of the rotating shaft;

add, over the first period of time, each of the first plurality of measurement values to a first local quantile sketch data structure, the first local quantile sketch data structure having a bounded physical size and being configured to maintain a summarization of the first plurality of measurement values;

send the first local quantile sketch data structure to the central computing device;

receive, from the central computing device, a second global combined quantile sketch data structure comprising the first local quantile sketch data structure and one or more other local quantile sketch data structures;

obtain, a plurality of new measurement values of the first item;

for each new measurement value of the plurality of new measurement values:

add the new measurement value to a second local quantile sketch data structure;

add the new measurement value to the second global combined quantile sketch data structure; and

query the second global combined quantile sketch data structure to determine whether the new measurement value is an anomalous measurement value; and

reconfigure, responsive to determining that at least one new measurement value is an anomalous measurement value, an operating behavior of the physical machine.

12. A system comprising:

a first computing device comprising:

a first memory; and

a first processor device coupled to the first memory to:

send, to a central computing device as the first computing device initiates, a request for a current global combined quantile sketch data structure comprising a plurality of local quantile sketch data structures received by the central computing device from a plurality of local computing devices;

obtain, over a first period of time, a first plurality of first item measurement values of a first item being measured, wherein the first item corresponds to a physical machine comprising a rotating shaft, and the first plurality of first item measurement values correspond to rotation speeds of the rotating shaft;

add, over the first period of time, each of the first plurality of first item measurement values to a first local quantile sketch data structure, the first local quantile sketch data structure configured to maintain a summarization of the first plurality of first item measurement values;

send the first local quantile sketch data structure to the central computing device;

receive, from the central computing device, a second global combined quantile sketch data structure comprising the first local quantile sketch data structure and a second local quantile sketch data structure generated by a second computing device;

obtain a plurality of new first item measurement values of the first item;

for each new measurement value of the plurality of new first item measurement values:

add the new measurement value to a second local quantile sketch data structure;

add the new measurement value to the second global combined quantile sketch data structure; and

query the second global combined quantile sketch data structure to determine whether the new first item measurement value is an anomalous measurement value; and

reconfigure, responsive to determining that at least one new first item measurement value of the plurality of new first item measurement values is an anomalous measurement value, an operating behavior of the physical machine.

13. The system of claim 12 wherein the first processor device is further to:

obtain a second plurality of first item measurement values of the first item;

add, to a second local quantile sketch data structure, each of the second plurality of first item measurement values;

obtain, by the first computing device, a second new measurement value of the first item; and

query a combination of the second global combined quantile sketch data structure and the second local quantile sketch data structure to determine whether the second new measurement value is an anomalous measurement value.

14. The system of claim 12 wherein the first processor device is further to determine, based on a timer, that the first local quantile sketch data structure is to be sent to the central computing device.

15. The system of claim 12 , further comprising:

a second computing device comprising:

a second memory; and

a second processor device coupled to the second memory to:

obtain, over the first period of time, a first plurality of second item measurement values of a second item being measured;

add, over the first period of time, each of the first plurality of second item measurement values to the second local quantile sketch data structure, the second local quantile sketch data structure configured to maintain a summarization of the first plurality of second item measurement values;

send the second local quantile sketch data structure to the central computing device;

receive, from the central computing device, the second global combined quantile sketch data structure comprising the first local quantile sketch data structure and the second local quantile sketch data structure;

obtain a new second item measurement value of the second item; and

query the second global combined quantile sketch data structure to determine whether the new second item measurement value is an anomalous measurement value.

16. The system of claim 15 further comprising:

the central computing device comprising:

a third memory; and

a third processor device coupled to the third memory to:

receive the first local quantile sketch data structure and the second local quantile sketch data structure;

combine the first local quantile sketch data structure and the second local quantile sketch data structure into the second global combined quantile sketch data structure; and

send the second global combined quantile sketch data structure to the first computing device and to the second computing device for use in determining measurement values that are not desired measurement values.