Planning for multiple Jazz application server instances

Authors: TimFeeney, RalphSchoon, MarianneHollier, DennisSchultz, StevenBeard
Build basis: The Rational solution for Collaborative Lifecycle Management (CLM) and the Rational solution for systems and software engineering (SSE)

The Jazz architecture allows for multiple instances of specific application servers, such as the Change and Configuration Management (CCM) application (Rational Team Concert), Quality Management (QM) application (Rational Quality Manager) and Requirements Management (RM) application (DOORS Next Gen). These application servers are registered to a common Jazz Team Server (JTS) to form a CLM Instance. The architecture further allows multiple CLM Instances to be formed. You should be aware of the tradeoffs and architectural considerations when multiple instances are needed in order to scale to support your user population/load.

Multiple Jazz Team Server considerations

It is possible to use multiple JTSes each with their own set of registered applications (each referred to as CLM Instance).

It is possible to distribute load to separate CLM instances e.g. one CLM instance per project or per department

Motivations

  • Maximize scalability
  • Prior to 5.0, JTS/RM shared storage leading to potential overtaxing of JTS due to RM’s Jena queries thus limiting number of concurrent users a single JTS/RM could handle; in such cases, multiple JTS are required to add another RM.

Pro:

  • Useful if projects are independent e.g.: separating subcontractors, compliance

Con

  • Administrative cost – increasing numbers need automation for deployment, administration and upgrade
  • More challenging to get a comprehensive view; if desired, would require Insight to report across multiple JTS

The image below shows multiple JTS and their registered applications. The reporting is working a cross all these CLM instances.
Multiple_JTS.png

Functional differences to be aware when multiple JTS exist

Reporting

  • RRDI reports only on data for applications associated to one JTS; multiple JTS/CLM Instances require JRS (Jazz Reporting Services) or Insight to report across JTS boundaries

Linking between Artifacts

  • Applications are registered to one JTS and it is possible to link artifacts between applications
  • Multiple JTS require manual creation of friends relationships to connect artifacts
  • Only CLM link types are available across application boundaries limiting choices of link types between artifacts

Cross-repository + cross-project link types

Link Name/Backlink Name Remote provider type
Related Change Request/Related Change Request CM
Affects Plan Item/Affected By Defect CM
Tracks/Contributes To CM (RTC only)
Tracks Change Set CM (RTC only)
Blocks Execution Record/Blocked By Change Request QM
Affects Execution Result/Affected By Change Request QM
Tested By Test Case/Tests Change Request QM
Related Test Case/Related Change Request QM
Related Execution Record/Related Change Request QM
Related Test Plan/Related Change Request QM
Implements Requirement/Implemented By RM
Affects Requirement/Affected By RM
Tracks Requirement/Tracked By RM
Elaborated By RM

Same project + Cross-project/Same repo link types

Link NameSorted descending Provider
Time Sheet Entry RTC
Resolves/Resolved By RTC
Related/Related RTC
Related Artifact/NA RTC
Predecessor/Successor RTC
Parent/Child RTC
Mentions/Mentioned By RTC
Duplicate Of/Duplicated By RTC
Copies/Copied From RTC
Change Set/NA RTC
Blocks/Depends On RTC
Attachment/NA RTC

Lifecycle Projects and CLM Instances

  • Cannot create lifecycle projects that include artifact containers registered with a different JTS

Dashboards live within one JTS

  • Can show widgets from applications with friends relationship

Multiple Jazz application server considerations

You can also deploy multiple instances of a Jazz application, such as the CCM application. If you deploy multiple instances of the same application in the same application server, you must give each instance a separate context root. For example, the context roots for two CCM instances could be ccm1 and ccm2. To connect multiple instances of the CCM application to a shared Jazz Team Server, the instances must all be authenticated from the same authentication realm and thus share the same set of users. In any deployment, the licenses are managed by Jazz Team Server. For more information about licenses, see Client access license management overview.

The image below shows multiple CLM applications registered to one JTS:
Multiple_CLM_instances.png

When you choose a topology for your deployment, carefully consider both the present and future needs of your team. While it is possible to move applications to a different application server later, this change requires the use of a proxy server to maintain links to that application.

Motivations

  • Capacity planning given the amount of users in the application instance and or usage model require multiple application servers
  • Funding model for projects requires segregation (e.g. not a shared, joint/commonly funded resource)
  • Isolate a department or subcontractor; separate confidential from non-confidential data - when OOTB permissions settings are insufficient

Pros

  • Balance load on applications
  • Increased scalability

Cons

  • Side effects (functional differences)
  • Increased reporting complexity

Multiple CCM Applications From a User Perspective describes how this would look like from a user perspective.

Multiple CCM Applications Best Practices describes some best practices to be considered when working in this environment.

Functional differences to be aware when multiple instances of Jazz Applications exist

Reporting

  • JRS supports single or multiple DW sources thus supporting multiple JTS scenario. It can also handle reports that need to cross application boundaries.

Linking between Artifacts

  • Applications are all registered to one JTS and it is possible to link artifacts between applications
  • Only CLM link types available across application boundaries limiting choices of link types between artifacts

Lifecycle Projects and CLM Instances

  • Projects in all applications can take part in lifecycle projects
  • Lifecycle Project templates don't have means to distinguish between multiple registered application instances of the same type, which might require to manually add projects to the lifecycle project

Installation considerations when deploying multiple instances of a Jazz application

Since each Jazz application instance must have a unique context root, when installing multiple instances of an application, do so from the command line and follow the instructions to Choosing a different context root than default

Multiple Change and Configuration Management application considerations

Functional differences to be aware of when multiple CCM instances exist

Work Items

  1. Work Items can be created in any CCM repository and linked to any other work item in any other repository
    • Selection dialog allows you to choose from repositories with a friends relationship to find or create work items
    • The only link relationships between work items in different CCM repositories are ‘Related Change Request’ and ‘Tracks’
    • To link work items, repositories need to be “Friends”
      • Automatic when registered to the same JTS
      • Set manually between applications in different CLM Instances
  2. Copying/moving of Work Items to another Project Area is only possible within the same CCM repository (see 169824)
  3. The Eclipse “current work item” functionality doesn’t work when the work item is in another repository

Planning

  1. Cross Project Planning
    • Cross project plans allow one relationship type to show as a tree view and roll up plan data
    • Must use the Tracks relationship to work across projects and repositories
    • Parent/child relationships are not supported across CCM repositories
    • Cross project plans show only the tracked work item and its rolled up plan data, parent child relationships underneath are hidden
  2. Work allocation
    • Users working in multiple CCMs will need to maintain their work allocation and scheduled absences in each CCM where they have assigned work. Correct and complete allocation and time off entry ensures proper load and progress calculations in plan views.

SCM

  1. Versioning code across CCM applications requires distributed SCM to replicate changes between CCM repositories
  2. Distributed SCM creates identical copies of change histories, change sets and changes made
    • Only the current baseline is migrated to the target repository
    • Select incoming baselines from the remote stream and accept the baselines to replicate them, if needed
    • Work Item links on change sets are replicated with the change set, but the work item only links back to the original change set
  3. To be compatible between repositories Distributed SCM requires both repositories to have the same metadata model thereby all RTC servers must be on the same version.
  4. Compatibility with previous versions is not supported in a distributed SCM setup. This means that there is no guarantee that distributed SCM works between different versions, as there could be comprehensive changes in how SCM manages its data between versions.
  5. Snapshots are local to a repository
    • No snapshot across multiple CCM repositories
    • Snapshots can’t be transferred, selecting a snapshot owner in a different repository not possible
    • Snapshots are not replicated
  6. It is possible to associate work items to change sets, regardless of the repository either side lives in
    • Navigation to work items works as usual, once connection is made
    • The link created using ‘associate work item’ only works within one repository
    • When associating a change set to a work item in another CCM repository, use drag and drop or the ‘associate change request’ gesture to make the association rather than ‘associate work item’
  7. Preconditions
    • Most out of the box deliver preconditions can only be satisfied by work item links. Some examples:
      • The precondition requiring a change set to be associated to an approved work item only works for work items in same CCM repository as change set
      • The precondition requiring a work item associated to the change set with its additional options only works for work items in same CCM repository as change set
      • The Precondition requiring an associated work item to match a query only works for work items in same CCM repository as change set
  8. Other SCM capabilities
    • The capability to locate change sets currently only works with local change set/work item relationships of the type ‘associate work item’

The image below shows distributed SCM across multiple CCM appliactions where the client is connected to multiple CCM applications and can work against multiple repositories:
Distributed_SCM.png

Build

  1. The "Included In Build" and "Reported Against Build" links in a Build Result can only refer to work items in the same CCM. These link types use item links, not URI links, so the references are to same repository only (see 222103).

Multiple Quality Management application considerations

  1. RQM is intended to reuse as many test assets as possible, to reduce maintenance and improve productivity of the test teams. Specifically:
    • Test cases
    • Test scripts
    • Test suites
    • Keywords
    • Test data
    • Test environments – platforms and parameters
    • Test schedules – timelines and iterations
    • Quality objectives, entry criteria, exit criteria
    • Categories and custom attributes
  2. This reuse is by reference and only within a QM project area
  3. Unlike CCM, there is no notion of creating associations between assets in different QM project areas, let alone QM instances
  4. Copying any of the assets creates a separate, editable copy, meaning for the same asset, you have to maintain the original and the copies in two (or more) places
  5. Copying should be kept to a minimum to maximize reuse within projects on the same QM instance

Functional differences to be aware of when multiple QM instances exist

  1. Duplicating test artifacts
    • The “Duplicate” capability enables you to replicate a single test artifact or an entire collection of artifacts
    • Only supports duplication within the source project area or to a project area on the same application instance.
    • You cannot use Duplicate to copy artifacts to a project area on another server.
    • However there is an ‘as-is’ command line RQM Copy Utility allowing copying of artifacts across servers.
  2. Built-in reporting and Rational Reporting for Developer Intelligence reports are limited in scope to one application server.

Multiple Requirements Management application considerations

As of 5.0, multiple RM instances per JTS is supported. In a 5.0 configuration with multiple RM instances, all JTS and RM servers must be at 5.0, that is, mixing 4.0.x and 5.0 RM server versions is not supported

Functional differences to be aware of when multiple RM instances exist

  1. Linking of RM artifacts between RM instances is supported in 6.0.6 for configuration management enabled projects; otherwise, the only supported link type is OSLC References.
  2. Link constraints are not supported for cross-server linking
  3. Embedding of artifacts is only supported on the same RM instance
  4. Glossary terms can only be referenced between components on the same server
  5. RM views can be constructed that include links to artifacts from local and remote RM instances. The same performance considerations and caveats apply as when including information from CCM and QM.
  6. When filtering a view based on artifact attributes on the target side of the link, this can only be performed on artifacts within the same server. When the artifact on the target side of the link is on another server, you are limited to filtering on the existence (or non-existence) of the link.
  7. Quick Search is limited to projects on the local RM instance
  8. For configuration management enabled projects:
    • Types can only be imported from components on the same server (instead use ReqIF to update types between servers)
    • Component templates are not shared across servers (You can download a project template from one server and upload it to another server)

Strategies for distributing projects across application servers

IBM clients currently fall into three distinct groups when determining their CLM application expansion approach or how to distribute projects across their application instances:

  • Shopping Bag
  • Moving Van
  • Field of Dreams

Shopping Bag

Overview – create one to many servers and fill up the servers with “similar” types of assets. If a server gets filled, then another server is on-boarded. The 'shopping bag' reference comes from the idea that when going to the market/grocery store, typically like things are put in the same bag, e.g. canned foods, boxed goods, cleaning solutions, frozen foods, tend to be kept separate from one another.

Projects organized/distributed by: Domain, Segment, Development Type, etc.

Benefits: scientific/proactive approach since it requires monitoring, logical since you are combining “similar” assets

Moving Van

Overview – an estimate is made at the beginning of the CLM deployment and a server is sized, with the anticipation that all data will fit and performance will continue to be good, no matter how CLM is used. The 'moving van' reference comes from the idea that when moving goods/furnishings from one's home, you typically try to get the largest moving van/truck possible to hold all your things with some room to spare and all things are loaded together.

Projects organized by: ad hoc, smorgasbord (a large mixture of many different things)

Benefits – estimates are done once and that is all until performance issues start

Field of Dreams

Overview – a hybrid of the other two approaches, plan for logical groupings of projects across multiple servers sizing each to allow for spare capacity, include monitoring to know when additional servers are needed.

Projects organized/distributed by: Domain, Segment, Development Type, etc.

Benefits – logical organization combined with one time estimation

Factors affecting Jazz application architecture

CCM Application Architectural Considerations

  1. Organizations which require rolled up reports leveraging OOTB RTC reporting should reside in the same CCM.
  2. Organizations may leverage Insight to report across projects residing in separate CCM’s.
  3. Organizations which require hierarchical work item relationships across project areas must reside in the same CCM.
  4. Consider separating areas which anticipate significant project area growth into different CCMs
  5. Logically organize CCM’s to limit sprawl
  6. Minimize use of distributed SCM where possible and group project areas which share code base in the same CCM.
  7. Choose an approach that will provide the most flexibility and scalability

QM Application Architectural Considerations

Separating artifacts into multiple project areas provides the ability to completely restrict access between testing teams. There is a place for this with highly regulated teams, secure projects and so on. If you have test artifacts that are highly sensitive and should only be viewable by a small number of people, those test assets should be contained within a separate RQM project area, regardless of the RQM server it is housed on. But also consider that you will be trading reuse and commonality for that independence.

We recommend that you partition your RQM project areas to take advantage of the reuse capabilities by:

  1. Determine which of your systems are highly integrated and keep those test artifacts in a single project area. We recommend that you use a single RQM project area for all test artifacts that are pertinent to a given application domain. For example, all testing artifacts that are germane to testing “System A” and its related systems would be in a single RQM project area. Moving to this system-centric model will enable you to reuse system-centric test environments, test cases, test scripts, etc. across releases and projects. Collecting and managing artifacts by sub-system can be achieved with the use of Categories. You can create a standard set of shared (public) filters in order to provide views of artifacts at the sub-system level. This enables you to keep system related artifacts together in one project while still enabling you to filter and report at the sub-system level.
  2. Use team areas to segregate test artifacts and govern security and permissions by role. Team areas enable you to set permissions on artifacts such that one team may have full permission over their artifacts but have only read permission to another team’s artifacts. Again, this may be an option that enables you to manage permissions at the sub-system level yet share common assets across all sub-systems.
  3. Understand how non-functional requirements – the most reusable requirements (usability, reliability, performance and supportability) and therefore the highly-reusable test cases associated with those non-functional requirements will be reused across multiple RQM project areas. Consider creating an enterprise catalog of non-functional tests from which you can derive system-specific non-functional tests.

If you believe you will need to copy test artifacts from one project area to another, create those project areas on the same RQM server to enable ease of copying. Copying from one RQM server to another RQM server is not currently supported in the web interface and must be accomplished via the copy utility.

In addition, if you believe you will want to report across project areas, house those project areas on the same RQM server. Rational Insight must be used for reports the cross RQM servers.

RM Application Architectural Considerations

  1. Organize projects on RM instances along some common affinity, like with like, e.g. common domains, systems.
  2. Projects should share common project template and to have a common/unified type system.
    • Create ‘Master Type System’ project from the template; leave unpopulated
    • Create and populate multiple projects from the same template
    • As changes to types occur, update ‘Master Type System’ project then import updates to populated projects to keep types in sync
    • Templates do not span RM instance boundaries, thus the recommendation to organize projects by affinity
  3. Projects requiring linking of artifacts using Built-in or Custom link types must be on same RM instance. Linking artifacts to projects across servers must be by Reference type. This restriction is removed in 6.0.6 for configuration management enabled projects.

When to add additional Jazz application servers

The decision to add additional CLM Instances or Jazz application servers is dependent upon the following:

  • Monitor 5 key factors
    • transaction of data between servers
    • growth of the database
    • license usage (concurrent usage & usage model)
    • JVM memory, heap, threads, etc.
    • CPU resources
  • Analyze and extrapolate trends from the collected data
    • compare to published sizing/ranges data
  • Decide and implement new Jazz application server(s) as needed
  • See Jazz Deployment Monitoring

Typical number of users supported per application instance

It is very difficult to pinpoint exactly how many users can be supported by a single instance of CCM or QM. There are a number of factors to consider such as the number of active users and what those users are doing.

It is not all that helpful to look at the total number of registered users. The number of actual concurrent users will typically be far lower than the total number of registered users. Some anecdotal data from internal IBM systems has indicated that only about 1% of registered users are active on average with peak usage over a several week period reaching about 3%.

Another factor to consider is what those users are doing. In CCM, for example, a developer clearly loads a system more than a contributor. SCM and build operations are expensive where work item management typically is not. QM users typically load a system less than a developer, although how much load a QM user puts on the system can vary considerably as well. A tester who is actively running tests, creating test results and generating reports will tax the system much more than a test manager creating and editing text in test plans.

IBM tests performance of each CLM release. To assist customers in their topology planning, understanding how many application instances they will need, we have documented our CLM Sizing Strategy. This strategy provides an expected range of concurrent users an application instance can be expected to support given a particular server configuration under a defined work load.

Related topics: Deployment planning: Where to start?

External links:

Additional contributors: RosaNaranjo

This topic: Deployment > WebHome > DeploymentPlanningAndDesign > PlanForMultipleJazzAppInstances
History: r31 - 2019-10-10 - 12:50:05 - TimFeeney
 
This site is powered by the TWiki collaboration platformCopyright © by IBM and non-IBM contributing authors. All material on this collaboration platform is the property of the contributing authors.
Contributions are governed by our Terms of Use. Please read the following disclaimer.
Dashboards and work items are no longer publicly available, so some links may be invalid. We now provide similar information through other means. Learn more here.