"Change Point Detection (#ChangeDetection) for MongoDB Time Series Performance Regression" paper for ACM/SPEC ICPE 2022 Data Challenge Track

UPDATE: the paper were published - (LINK to PAPER)

The ACM/SPEC ICPE 2022 - Data Challenge Track Committee has decided to ACCEPT our article:

TITLE: Change Point Detection for MongoDB Time Series Performance Regression 
AUTHORS: Md Shahriar Iqbal, Mark Leznik, Igor Trubin, Arne Lochner, Pooyan Jamshidi and André Bauer

https://icpe2022.spec.org/tracks-and-submissions/data-challenge-track/

ABSTRACT

Commits to the MongoDB software repository trigger a collection

of automatically run tests. Here, the identification of commits 

responsible for performance regressions is paramount. Previously, the

process relied on manual inspection of time series graphs to identify

signi￿cant changes, later replaced with a threshold-based detection

system. However, neither system was sufficient for finding changes

in performance in a timely manner. This work describes our recent

implementation of a change point detection system built upon the

Perfomalist approach in combination with XGBoost algorithm. The

algorithm produces a list of change points representing significant

changes from a given history of performance results. We are able

to automatically detect change points and achieve an 83% accuracy,

all while reducing the human effort in the process.

More Perfomalist's  approach details can be found in this blog post:

CPD - Change Point Detection (#ChangeDetection) is implemented in the free web tool Perfomalist

The result of initial usage of Perfomalist CPD API against MongoDB data is published HERE:

Perfomalist #ChangeDetection API was used against #MongoDB #perfomanceTesting dataset

 

 

My Cloud Optimization team at #CapitalOne bank won the CMG.org #Innovation Award (#CMGNews)

  https://www.cmg.org/2022/02/capital-one-announced-as-winner-of-the-impact-innovation-award/

My publications in RG got 5000+ reads

https://www.researchgate.net/profile/Igor-Trubin 

Panel Discussion: Roadmap for Cultivating Performance-Aware Software Engineers

 

"#CloudServers Rightsizing with #Seasonality Adjustments" - my presentation at CMG IMPACT conference (#CMGnews)

Feb 4, 2022 12:15 Virtual at https://cmgimpact.com/sessions-schedule/

"Performance Anomaly and Change Point Detection for Large-Scale System Management" - my paper published at Springer

 

Intelligent Sustainable Systems pp 403-407| Cite as

Performance Anomaly and Change Point Detection for Large-Scale System Management

• Authors
• Authors and affiliations

• Igor Trubin

1. 1.

Conference paper

First Online: 17 December 2021

• 1Downloads

Part of the Lecture Notes in Networks and Systems book series (LNNS, volume 334)

Abstract

The presentation starts with the short overview of the classical statistical process control (SPC)-based anomaly detection techniques and tools including Multivariate Adaptive Statistical Filtering (MASF); Statistical Exception and Trend Detection System (SETDS), Exception Value (EV) meta-metric-based change point detection; control charts; business driven massive prediction and methods of using them to manage large-scale systems such as on-prem servers fleet or massive clouds. Then, the presentation is focused on modern techniques of anomaly and normality detection, such as deep learning and entropy-based anomalous pattern detections.

Keywords

Anomaly detection Change point detection Business driven forecast Control chart Deep Learning Entropy analysis 

This is a preview of subscription content, log in to check access.

References

1. 1.
   Trubin, I.: Exception based modeling and forecasting. In: Proceedings of Computer Measurement Group (2008)Google Scholar
2. 2.
   Jeffrey Buzen, F., Annie Shum, S.: MASF—multivariate adaptive statistical filtering. In: Proceedings of Computer Measurement Group (1995)Google Scholar
3. 3.
   Trubin, I.: Review of IT control chart. CIS J. 4(11), 2079–8407 (2013)Google Scholar
4. 4.
   Perfomalist Homepage, http://www.perfomalist.com. Last accessed on 10 June 2021
5. 5.
   Trubin, I., et al.: Systems and methods for modeling computer resource metrics. US Patent 10,437,697 (2016)Google Scholar
6. 6.
   Trubin, I.: Capturing workload pathology by statistical exception detection. In: Proceedings of Computer Measurement Group (2005)Google Scholar
7. 7.
   Loboz, C.: Quantifying imbalance in computer systems. In: Proceedings of Computer Measurement Group (2011)Google Scholar

Dynamics of Anomalies or Phases in a Dynamic Object Life

A dynamic object may have following several phases in its lifetime:

1.  Initial phase to set a norm - anomalies cannot be detected as there is no baseline sample is established yet. Could be tired later as an outlier.

2.  Stable period without any anomalies.

3. Unstable period when anomalies are appearing: suddenly or with gradually increasing rate.

4. Anomalies are introducing a new norm and the rate of anomalies is gradually decreasing.

5. =>2. The next stable period. 

6. =>3. … and so on.

To detect those dynamic object phases one can use Anomaly and Change Point detection methods. One of them is SETDS (described in this blog), which has been implementing now as a www.Perfomalist.com tool. 

Here is an example how the Perfomalist (Download Input Data Sample) test data is used to detect stable and unstable periods.

Data consists of 28 weeks. To see some dynamic and to  catch when anomalies started appearing, the data was divided into 23 data sets. 

- The 1st one has 4 initial weeks (initial baseline or reference/learning set) plus following week (1st "current" week). 

- The 2nd one has 5 initial weeks as the next (on one week bigger) baseline and following week as the next  "current" week. 

- The 3rd one... the same mechanism as described above.

Then the www.Perfomalist.com was applied 23 times (could be automated using Personalist APIs)  and results were combined into the spreadsheet. 

The table and daily summarized charts are below. The result shows clearly 2nd (stable)  and 3rd (unstable) phases. 

Another way to detect is CPD and to do that another Perfomalist API can be used.

Applying that to the same data the similar result is seen:

The results are similar but a bit different and I know why.... 

Join me with CMG – your technology community – at #CMGIMPACT22. Use code Trubin at cmgimpact.com/ for 50% off IMPACT tickets cmgimpact.com/register/ #cmgnews #technology #InformationTechnology #ITconference #ContinuingEducation #ProfessionalDevelopment

PLEASE REGISTER: https://cmgimpact.com/register/

https://cmgimpact.com/cloud-servers-rightsizing-with-seasonality-adjustments/ 

When the cloud servers rightsizing algorithm calculates the baseline level for the current year application server’s usage, the seasonal adjustment needs to be calculated and applied by adding the additional anticipated change, which could be increasing or decreasing the capacity usage. We describe the method and illustrate it against the real data.

The cloud servers rightsizing recommendation generated based on seasonality adjustments, would reflect the seasonal patterns, and prevent any potential capacity issues or reduce an excess capacity.
The ability to keep multi-year historical data of 4 main subsystems of application servers’ capacity usage opens the opportunity to detect seasonality changes and estimate additional capacity needs for CPU, memory, disk I/Os, and network. A multi-subsystem approach is necessary, as very often the nature of the application could be not CPU but I/Os or Memory or Network-intensive.

Applying the method daily allows downsizing correctly if the peak season passes and the available capacity should be decreased, which is a good way to achieve cost savings.

In the session, the detailed seasonality adjustment method is described and illustrated against the real data. The method is based on and developed by the author’s SETDS methodology, which treats the seasonal variation as an exception (anomaly) and calculates adjustments as variations from a linear trend.

Key Takeaways

• How to build seasonal adjustments into the cloud rightsizing
• To get familiar with cloud objects rightsizing techniques

The Change Point Detection SETDS based method is implemented as a Perfomalist API. Everybody is welcome to test!

How to use it explained HERE:

https://www.trutechdev.com/2021/11/the-change-points-detection-perfomalapi.html

Example of the step jump event detected by the API (Output got from the API call via Postman and spreadsheet was used to chart the result):

My presentation "Cloud Servers Rightsizing with Seasonality Adjustments" has been accepted for CMG IMPACT 2022. #CMGnews

www.CMGimpact.com 

Got my 1st #AWScertification

Update 2023:

I have got also:

AWS solution architect (associate)AWS solution architect (associate)

Amazon Web Services (AWS)Amazon Web Services (AWS)Issued Aug 2022Issued Aug 2022

Credential ID Z0R8ZQH1Y2QQ11S5  

View of  my verified achievement from Amazon Web Services (AWS) is HERE

AWS Certified Cloud Practitioner was issued by Amazon Web Services Training and Certification to Igor Trubin

"Performance #Anomaly and #ChangePointDetection For Large-Scale System Management" for WorldS4 2021 - my presentation slides deck is available on RG

 I have successfully made my presentation at WorldS4 conference. Presentation deck is available HERE

I'm excited to present my paper "Performance #Anomaly and Change Point Detection for Large-Scale System Management" at 5th World Conference on Smart Trends in Systems, Security and Sustainability

See that in the agenda: https://sched.co/lEkJ

 

Presenting in London - "Performance Anomaly and Change Point Detection For Large-Scale System Management"

I will be presenting at the Worlds3 conference in London my paper

"Performance Anomaly and Change Point Detection For Large-Scale System Management"

(https://www.researchgate.net/publication/340926055_Performance_Anomaly_and_Change_Point_Detection_For_Large-Scale_System_Management)

Time slot in London time: 04:30 - 06:00 on 29th July 2021

Cloud Capacity Management Explained by CMG.org - #cmgnews

CMG publications about Cloud Capacity Management (some links accessible only for CMG members)

1. Cloud Capacity Management (PDF doc from Metron-Athene)

2. 8 Things You Need to Know About Capacity Planning for the Cloud (helpsystem)

3. How to Do Capacity Management in the Cloud (helpsystem).

4. (in UT) How to do Capacity Management in the Cloud Text is HERE (TeamQuest)  

5. Building and Rebuilding a Data Center Every Day  (Netflix)

6. Netflix Performance Tales in One Take

7. Cloud Cost Optimization at Spotify

8. Lifting the Cloud of Obscurity from your Cloud Deployment

9. The new dimensions of cloud – IT infrastructure resource planning, optimization and cost BMC Software

10. Redefining Enterprise Cloud Transformations: How Fidelity Investments is establishing a new foundation for Observability and Reliability (is coming)

11. Cloud Capacity Management  by Kevin McLaughlin (Capital One)

12. Under cloudy skies capacity planning in changing times – Brian Wong, Capital One

13. Optimizing your Cloud – Igor Trubin, IT Manager, Capital One

How am I doing? LinkedIn recommendations (this year)

 

See more - https://www.linkedin.com/in/itrubin/

SEDS based "CLOUD RESOURCES WORKLOAD PROFILING"

Based on SEDS method the workload profiling of main cloud objects (AWS EC2, RDSand  EBS) are implemented at my current work. 

Next Tuesday 3/30 at 12:30 pm EST I will be sharing my experience of building and using this method at the Data Centers and Cloud Infrastructure virtual CMG.org conference. You are welcome! The topic of the presentation is "CLOUD RESOURCES WORKLOAD PROFILING"

ABSTRACT: How to be sure a cloud object’s (e.g, AWS EC2, RDS or EBS) workload fits the rightsized resources (Compute, RAM, IO/s and Network traffic)? It is very difficult to do using raw system performance data from monitoring tools. The best way to do that is using a weekly workload profile, which is a graphical visualization in form of MASF IT-Control chart. This chart shows the stability of the workload, reveals the anomalies that happened recently, such as run-away, memory leaks or specifically important for cloud objects, the unusual number of hours the object is down all compared with the usual weekly pattern.

This presentation will describe how to build, read, and use workload profiles using real data examples and demonstrates how cloud capacity scaling could be verified.