Users may want to improve the processing performance of queries by allowing queries to split into multiple processing threads.
Multi-threaded queries allow IBM Cognos TM1 to automatically load balance a single query across multiple cores. This can improve efficiency and processing time for large queries and rules.

Previous Customer Concerns:

• CPU Utilization: “I’ve got 16 cores and my CPU utilization is at 15%”
• Server PVU Value: “More cores do not make my queries faster”
• Data Scale: “TM1 Solutions have a data volume ceiling”
• Rule Caution: “Rules slow down my queries to an unacceptable performance level”

• Simple Configuration: tm1s.cfg -> MTQ =
• All UIs can leverage MTQ: TM1 multi-threads stargate cache creation
• High Performance: Query speed improves relative to available cores
• Manages Concurrency: Available cores are load balanced across queries

The MTQ setting is part of overall Performance tuning in TM1 configuration (cfg) file:

• MTQ (default 0)
• MaximumCubeLoadThreads (default 0)
• PersistentFeeders (default F)
• ParallelInteraction (default T)
• UseLocalCopiesForPublicDynamicSubsets (default T)
• ViewConsolidationOptimization (default T)
• ViewConsolidationOptimizationMethod (default ARRAY)

• Power Plan setting High Performance- not balanced
• Enable Hyperthreading (BIOS setting in some cases)
• Optimize the use of the TM1 Query cache via configuration of the VMM/VMT parameter (CubeProperties) (see our accompanying blog post!)

MTQ is applied to queries exceeding 10,000 cell visits
Creation of TM1 ‘stargate’ cache is multi-threaded
Available server cores are applied to concurrently processing queries
MTQ automatically load balances cores across concurrent queries

• The query is parsed into worker threads, each performing their own transaction
• Large complex (rule heavy) views will see large gains, but often not linear to the number of cores assigned.
• Large non-complex (rule light or no rules) views see exponential gains, often linear to the number of cores assigned

• Assumption : 8 core server, MTQ = 8
• User 1 launches large data query:
• Assigned 1 master thread, and 7 worker threads
• User 2 launches second query:
• Query 2 is queued and gets assigned 4 threads, leading to a 4/4 split between users 1 and 2
• User 3 launches third query:
• Query 3 is queued and gets assigned 2 threads, Queries 1 and 2 continue on 3 threads

Test results from the IBM Labs have revealed impressive performance gains for MTQ versus Single-Threaded Queries.

The impact of additional cores on query times with MTQ implemented for a Large TM1 Model Customer were investigated as below.

Customer Model Overview:

• 75G Model, cubes 3G – 10G (Model Size)
• 20 Concurrent Users (# of users)
• 64 Core Server, 512G RAM (Ram and Server Size)

This is attributed to the fact that the non-stargate operations associated with displaying a view still require a fixed amount of time.

• Memory considerations
  • MaximumCubeLoadThreads can have a significant impact on Memory consumption due to duplicate feeders
  • MTQ setting also has an impact on memory consumption, but to a lesser extent
• IBM recommendations regarding the number of cores
  • MTQ: maximum number of cores available
  • MaximumCubeLoadThreads:  set to 50% of total cores (old recommendation was max-1)

The performance gains of MTQ can be leveraged within TurboIntegrator by calling a sub-process that will create a cached view using the ‘ViewConstruct’ TI function.