Presentation is loading. Please wait.

Presentation is loading. Please wait.

Dr. Spotfire: Scaling Spotfire

Similar presentations


Presentation on theme: "Dr. Spotfire: Scaling Spotfire"— Presentation transcript:

1 Dr. Spotfire: Scaling Spotfire
Peter McKinnis Architect, Analytics Enterprise Architect Group December 3, 2019

2

3

4 © Copyright 2000-2019 TIBCO Software Inc.
About TIBCO TIBCO fuels digital business by enabling better decisions and faster, smarter actions through the TIBCO Connected Intelligence Cloud. From APIs and systems to devices and people, we interconnect everything, capture data in real time wherever it is, and augment the intelligence of your business through analytical insights. Thousands of customers around the globe rely on us to build compelling experiences, energize operations, and propel innovation. Learn how TIBCO makes digital smarter at  © Copyright TIBCO Software Inc.

5 © Copyright 2000-2019 TIBCO Software Inc.
Architecture review Scaling considerations Spotfire Server Web Player Helpful links © Copyright TIBCO Software Inc.

6 What to get out of presentation?
I hope you… Learn something new Get some pointers to resources that will be helpful Get some strategies and information to help you size and scale your environments What you will not get Some secret magic formula for computing the exact scale and size of an environment Go to Monitoring & Diagnostics and Action Logs and Performance Counter Example © Copyright TIBCO Software Inc.

7 TIBCO Spotfire Architecture
Spotfire Analyst Consumer and Business Author Spotfire Server Web Player Mobile Access © Copyright TIBCO Software Inc.

8 Scaling Considerations - Users
Types of users Number of concurrent users of each type Or Total number of users and can estimate concurrent (concurrent is typically 10-20% of total) Business Author Analyst Consumer Spotfire Server Web Player Spotfire Server © Copyright TIBCO Software Inc.

9 Scaling Considerations - Data Size
Amount of RAM used by Spotfire analysis with data loaded Estimated Number of rows and columns of data Estimated database size of datasets Relative size can be useful. For example: 2X-Large > 200 GB X-Large GB Large 10’s GB Medium 1-10 GB Small 100MB-1GB X-Small < 100 MB © Copyright TIBCO Software Inc.

10 Scaling Considerations - Data Consumption
In-Memory In-Memory Load data from source into memory. External Data Leave data in database. Dynamically load and discard data to visualize. External On-demand On-Demand Dynamically swap data in and out of memory. Streaming Data Streaming Data Real-time data loaded into memory and out of memory as time window shifts. © Copyright TIBCO Software Inc.

11 Scaling Considerations – In-memory
If already have Spotfire analysis available, one can estimate RAM using Spotfire Analyst Open Spotfire Analyst and look at memory used by Spotfire.Dxp.exe in Task Manager: Open Spotfire Analysis in Spotfire Analyst and look at memory used by Spotfire.Dxp.exe in Task Manager: Consider formula of data size + user memory (typically MB – depending on complexity of analysis) This file - 1 Million rows x 100 Cols, about 70% character and 30% numeric columns - uses about 1.9 GB = GB GB Simple formula: Size_of_data + (user_data * # users) = MemoryNeeded – user_data typically between 40-80MB depending on analysis complexity. For example: (60 MB * 100) = 7.9 GB © Copyright TIBCO Software Inc.

12 Scaling Considerations – Data in-memory
# of Rows # of Columns # of Cells (Rows * Columns) # of Numeric Columns # of Character Columns Size in-memory (GB) Table size in SQL Server (GB) 1,000,000 100 30,000,000 30 70 1.95 1.6 1,500,000 9 13,500,000 2 7 0.6 0.190 10,000,000 40 400,000,000 25 14 5.9 4.0 15,000,000 135,000,000 7.5 1.9 50,000,000 20 1,000,000,000 10 13.9 13.1 In-memory data size affected by: entropy of data (a few or many unique values in a column) data types of columns (character, numeric, number of characters)

13 Scaling Considerations – Spotfire Analyses
Number of Analyses What is largest analysis in-memory? Complexity of Analyses: Calculations - calculated columns, custom expressions Scripts - IronPython, TERR, etc. Can affect CPU load Is analysis pre-loaded by Scheduled Updates? This saves a lot of CPU by not having to load the file for each user. Recommended for highly used analyses and important analyses that take a while to load Is data shared by users? Reduces memory footprint in Web Player How to setup caching for reports with row-level security using personalized Information Links © Copyright TIBCO Software Inc.

14 © Copyright 2000-2019 TIBCO Software Inc.
When to scale? New environment: Use information regarding – users, data, etc. to scale environment Consider - How precise do I need to be in my sizing? Can I start small and scale up as needed or do I need to know how big for the next 3 years? Changes in environment: New Spotfire Analyses being deployed Increase in Spotfire users Increase in amount of data in-memory Increase in number of on-demand queries Need to support high availability Addition of Streaming data Performance issues in environment: Need to determine where the performance issue might be (could be external issue) Is hardware at capacity – memory, CPU, disk? Current view in Monitoring & Diagnostics Administration Web UI Historical information available from: Action Logs and System Monitoring Performance Counter Log Files Go to Monitoring & Diagnostics and Action Logs and Performance Counter Example © Copyright TIBCO Software Inc.

15 TIBCO Spotfire Server - Scaling
Items most related to scaling Spotfire Server: Number of concurrent data connections via Information Services (on-demand and other queries) Number of concurrent active Consumers/Business Authors and, to some extent, Analyst users Need for high availability Streaming data used on Web Player (10.6 and above) – increases amount of HTTP traffic since visuals constantly updating What to scale? Scale up - Additional memory and CPUs Increase JVM Memory for Spotfire Server In Spotfire Server configuration, Increase max-jobs and thread-pool-size (Information Services jobs) Scale out - Additional Spotfire Servers Use same Spotfire database so install server and configure to use existing database Need load balancer with sticky sessions enabled in front of 2 or more Spotfire Servers Spotfire Servers tend to not be too large, e.g. 4-8 cores and GB RAM Show how to increase JVM RAM on 10.3.x © Copyright TIBCO Software Inc.

16 TIBCO Spotfire Web Player Service – Scaling
Items most related to scaling Web Player Service: Number of concurrent active Web Player users Largest Analysis file needed in-memory Complexity of calculations in analysis Sharing of data in analysis Use of Scheduled Updates How data will be consumed: In-memory External Data on-demand Streaming added in 10.6 for Web Player (See Scaling streaming visual analytics with Spotfire Web Player) What to scale? Scale up Additional memory and CPUs If scale up very large, additional Web Player instances on the same node provides more throughput. Scale out - Additional Web Player Nodes Web Player Nodes tend to larger than Spotfire Server, e.g cores and GB RAM © Copyright TIBCO Software Inc.

17 © Copyright 2000-2019 TIBCO Software Inc.
Helpful Links Web Player Performance Information: Accessing performance data of Services Web Player service performance counters Troubleshooting Performance issues on TIBCO Spotfire Web Player Streaming data - Spotfire 10.6 and later - Scaling streaming visual analytics with Spotfire Web Player Spotfire Server Configuration: Spotfire Server hangs or responds slowly with error: “Too many jobs” Spotfire Server JVM Memory Settings: Modifying the virtual memory 10.3.x TIBCO Spotfire Server in a cluster crashes due to long GC pauses © Copyright TIBCO Software Inc.

18 © Copyright 2000-2019 TIBCO Software Inc.


Download ppt "Dr. Spotfire: Scaling Spotfire"

Similar presentations


Ads by Google