Message queues at Salesforce allow for asynchronous processing of jobs like dashboards, reports, and bulk API calls. Salesforce sees around 60 million messages per day across instances, with peak instances processing 10 million daily. The queues use a large, shared thread pool for round-robin processing to reduce wait times, allocate resources fairly, and adapt to varying traffic patterns. Elastic thread pools allow the system to grow and shrink processing capacity based on current demand and wait times.

1. Message Queues at salesforce.com
The subtitle goes here
Vijay Devadhar
Developer
Salesforce.com

2. Safe Harbor
Safe harbor statement under the Private Securities Litigation Reform Act of 1995:
This presentation may contain forward-looking statements that involve risks, uncertainties, and assumptions. If any such uncertainties materialize or if
any of the assumptions proves incorrect, the results of salesforce.com, inc. could differ materially from the results expressed or implied by the forward-
looking statements we make. All statements other than statements of historical fact could be deemed forward-looking, including any projections of
product or service availability, subscriber growth, earnings, revenues, or other financial items and any statements regarding strategies or plans of
management for future operations, statements of belief, any statements concerning new, planned, or upgraded services or technology developments
and customer contracts or use of our services.
The risks and uncertainties referred to above include – but are not limited to – risks associated with developing and delivering new functionality for our
service, new products and services, our new business model, our past operating losses, possible fluctuations in our operating results and rate of growth,
interruptions or delays in our Web hosting, breach of our security measures, the outcome of intellectual property and other litigation, risks associated
with possible mergers and acquisitions, the immature market in which we operate, our relatively limited operating history, our ability to expand, retain,
and motivate our employees and manage our growth, new releases of our service and successful customer deployment, our limited history reselling
non-salesforce.com products, and utilization and selling to larger enterprise customers. Further information on potential factors that could affect the
financial results of salesforce.com, inc. is included in our annual report on Form 10-Q for the most recent fiscal quarter ended July 31, 2012. This
documents and others containing important disclosures are available on the SEC Filings section of the Investor Information section of our Web site.
Any unreleased services or features referenced in this or other presentations, press releases or public statements are not currently available and may
not be delivered on time or at all. Customers who purchase our services should make the purchase decisions based upon features that are currently
available. Salesforce.com, inc. assumes no obligation and does not intend to update these forward-looking statements.

3. Vijay Devadhar
Developer

4. Message Queues – what are they?
 Asynchronous job queue infrastructure in salesforce.com server
 Engine behind
• Dashboards
• Reports,
• Batch/Async Apex
• Bulk API
• and many more…

5. Message Queues – what are our volumes?
 Averages about 60 million messages a day
 Biggest instances account for 10 million messages a day
 95 percentile for dequeue latency is 10 minutes

6. Message Queues – session description
 Discuss scaling techniques used to
• Manage capacity
• Resource allocation
 and lessons learned..

9. Visit with Princess Aurora
 Average wait time – 90 mins
 Average time with princess – 120 seconds
 What you remember the most after a year? The long……… wait. 

11. Visit with Goofy
 Average wait time – none
 Average time with goofy – 120 seconds or until you get bored

12. Meanwhile…..

13. Characters are not Created Equal

14. If Only We Hired More Princess Aurora

15. What can we do better?
 Fire Goofy(s)
 Hire more Princesses
 Convert Goofy to Princess and vice-versa when needed

16. Swap Goofy for Princess Aurora

18. Message Queues Amusement Park
 300 + rides and characters
 Traffic which ebbs and flows with time of day, day of the week etc.,
 Plenty, but finite set of resources available

19. Goal
 Reduce wait times
 Fairly allocate resources
 Adapt to varying traffic patterns

20. Solution
 A large Shared thread pool – No ride specific silos
 Round robin the process of picking work
 If world wants Dashboards, do Dashboards

22. Message Queue Real Time Latencies
 Unlike Disneyland, each job takes variable amount of time
 Wait time prediction is not accurate at the tail of the queue
 We report real time and act on them if needed

25. Elastic Thread Pools
 Can grow from initial size
 This allows growth as traffic demands
 Wait times feedback to thread pool grow, shrink decisions

26. Let’s Do a Puzzle
 A man has
• One Sheep
• One Tiger
• One bundle of Grass
• One small boat
 and a big River to cross…

28. Similar Puzzle with Messages
 Forecasting has
• Several sales reps whose forecasts need update
• Forecast update for sales rep should also update VP of sales
• Multiple sales reps for same VP
 and a big sales projection to put out ….

29. Solution
 Browse and Cache
 Pick up work on which you can obtain mutex lock
 Jump ahead if needed

30. And in Real World..
 Cache capacity is tuned to typical traffic pattern
 At times cache can fill up
 Messages may be escorted to the back of the queue

31. Bread Lines Vs. Turkey Lines
 Same set of ovens baking both
 Bread is the basic need, Turkey when ovens are free
 If bread lines build up, stop cooking turkey
 If no one wants bread, just give all ovens to cooking turkey

32. User-facing vs Background jobs
 Same set of servers for both
 Users need fast response; Background can wait
 If user requests pile up, stop processing background
 If no user requests, just process background jobs

33. But how do you make Turkeys stop?
 Traffic lights
 Measure key resources
 When resource usage crosses threshold, slow down on background
 Sensitive to CPU, Memory, I/O, connection usage

34. Lessons in Real World
 Traffic types vary, traffic volumes vary
 Handlers misbehave, components have bugs
 Distributed systems scale very well, not if you need mutex.
 Real time alerting, trending, traffic isolation, troubleshooting are
necessary