WARNING: This is the _old_ Lustre wiki, and it is in the process of being retired. The information found here is all likely to be out of date. Please search the new wiki for more up to date information.

Lustre Test Plans: Difference between revisions

From Obsolete Lustre Wiki
Jump to navigationJump to search
No edit summary
Line 28: Line 28:
Test plans are available for upcoming Lustre releases.
Test plans are available for upcoming Lustre releases.


=== Lustre 1.8 Test Plan ===
==== Lustre 1.8 Test Plan ====


This is the [http://wiki.lustre.org/images/6/68/Lustre-18-release-tp.pdf Lustre 1.8 Test Plan]. The objective is to find and resolve defects in the Lustre code during release testing cycles, with an ultimate goal of delivering production-ready Lustre code.
This is the [http://wiki.lustre.org/images/6/68/Lustre-18-release-tp.pdf Lustre 1.8 Test Plan]. The objective is to find and resolve defects in the Lustre code during release testing cycles, with an ultimate goal of delivering production-ready Lustre code.


=== Lustre 2.0 Test Plan (coming soon) ===
==== Lustre 2.0 Test Plan (coming soon) ====


==Feature Testing==
==Feature Testing==
Line 38: Line 38:
This section lists the feature test plans used by the Lustre QE team.
This section lists the feature test plans used by the Lustre QE team.


===Adaptive Timeouts===
====Adaptive Timeouts====


This is the [http://wiki.lustre.org/images/c/c7/AT-large-scale-tp.pdf Large-Scale Test Plan for Adaptive Timeouts]. The goal is to test the Adaptive Timeouts feature on a large-scale system.
This is the [http://wiki.lustre.org/images/c/c7/AT-large-scale-tp.pdf Large-Scale Test Plan for Adaptive Timeouts]. The goal is to test the Adaptive Timeouts feature on a large-scale system.


===CMD===
====CMD====


This is the [http://wiki.lustre.org/images/4/4a/CMD-large-scale-tp.2008-11-20.pdf Large-Scale Test Plan for CMD]. The objective is to test the Clustered MetaData (CMD) feature at scale, and verify that performance improvements are seen with CMD, as compared to a single MDT configuration.
This is the [http://wiki.lustre.org/images/4/4a/CMD-large-scale-tp.2008-11-20.pdf Large-Scale Test Plan for CMD]. The objective is to test the Clustered MetaData (CMD) feature at scale, and verify that performance improvements are seen with CMD, as compared to a single MDT configuration.
Line 48: Line 48:
NOTE: CMD is designed for use only on '''non-production''' Lustre systems.
NOTE: CMD is designed for use only on '''non-production''' Lustre systems.


===OSS Read Cache===
====OSS Read Cache====


This is the [http://wiki.lustre.org/images/8/82/ROC-large-scale-tp-1_021309.pdf Large-Scale Test Plan for OSS Read Cache]. The goal is to test OSS Read Cache on a large-scale system to be sure the feature scales, and verify that performance data, when OSS read cache is enabled, is better than data when the feature is disabled.
This is the [http://wiki.lustre.org/images/8/82/ROC-large-scale-tp-1_021309.pdf Large-Scale Test Plan for OSS Read Cache]. The goal is to test OSS Read Cache on a large-scale system to be sure the feature scales, and verify that performance data, when OSS read cache is enabled, is better than data when the feature is disabled.


===OST Pools===
====OST Pools====


This is the [http://wiki.lustre.org/images/e/e1/Ostpools-large-scale-tp_020209.pdf Large-Scale Test Plan for OST Pools]. The objective is to test OST Pools on a large-scale cluster to make sure the feature is scalable, and verify that file creation performance is the same (with or without OST Pools).
This is the [http://wiki.lustre.org/images/e/e1/Ostpools-large-scale-tp_020209.pdf Large-Scale Test Plan for OST Pools]. The objective is to test OST Pools on a large-scale cluster to make sure the feature is scalable, and verify that file creation performance is the same (with or without OST Pools).


===Quotas===
====Quotas====


This is the [http://wiki.lustre.org/images/7/75/Quotas_testplan.pdf Test Plan for Lustre Quotas]. The test plan's objective is to:
This is the [http://wiki.lustre.org/images/7/75/Quotas_testplan.pdf Test Plan for Lustre Quotas]. The test plan's objective is to:
Line 71: Line 71:
*lustre quotas feature code
*lustre quotas feature code


===Lustre Security===
====Lustre Security====


This is the [http://wiki.lustre.org/images/3/35/SecurityTestPlan.pdf Test Plan for Security Features]. The test plan's objective is to find defects in the Lustre 2.0 security features and drive the security source code to a stable, production-ready state. The test plan contains:  
This is the [http://wiki.lustre.org/images/3/35/SecurityTestPlan.pdf Test Plan for Security Features]. The test plan's objective is to find defects in the Lustre 2.0 security features and drive the security source code to a stable, production-ready state. The test plan contains:  
Line 79: Line 79:
* Complete I/O benchmark (IOR and mdsrate), to run against all Lustre-supported security flavors (null, plain, krb5n, krb5a, krb5i and krb5p), gather and analyze performance data.
* Complete I/O benchmark (IOR and mdsrate), to run against all Lustre-supported security flavors (null, plain, krb5n, krb5a, krb5i and krb5p), gather and analyze performance data.


===Version-Based Recovery===
====Version-Based Recovery====


This is the [http://wiki.lustre.org/images/f/f4/VBR_phase2_large_scale_testplan.pdf Large-Scale Test Plan for Version-Based Recovery]. The test plan's objective is to verify that the VBR feature works with a large system.
This is the [http://wiki.lustre.org/images/f/f4/VBR_phase2_large_scale_testplan.pdf Large-Scale Test Plan for Version-Based Recovery]. The test plan's objective is to verify that the VBR feature works with a large system.
Line 87: Line 87:
This section lists regression test plans used by the Lustre QE team.
This section lists regression test plans used by the Lustre QE team.


===Lustre 1.8 Performance Regression Test Plan===
====Lustre 1.8 Performance Regression Test Plan====


This is the [http://wiki.lustre.org/images/9/9a/Lustre_performance_regression-tp_013009.pdf Lustre 1.8 Performance Regression Test Plan]. The goal is to test each Lustre 1.8 feature for performance regression on an at-scale cluster.
This is the [http://wiki.lustre.org/images/9/9a/Lustre_performance_regression-tp_013009.pdf Lustre 1.8 Performance Regression Test Plan]. The goal is to test each Lustre 1.8 feature for performance regression on an at-scale cluster.

Revision as of 18:21, 24 April 2009

Test plans for upcoming Lustre releases are now available for review and comment. As an open-source product, we invite the Lustre user community to contribute to our efforts to develop and test a more robust, feature-rich Lustre.

Commenting on the Test Plans

If you want to comment on a test plan, click the Discussion tab at the top of the page and enter:

  • Your name and email address
  • Date of your comment
  • Test plan you are commenting on (e.g., 1.8, VBR, etc.) and page number, if applicable
  • Comment or suggestion. If you have a proposed test case, paste it into your comment.

The Lustre Test group will review these comments, update test plans as needed, and post updated plans to this page.

Feature Testing with the Test Plans

If you want to try out a feature in your Lustre environment, use the appropriate feature test plan to verify the functionality. We especially need help with large-scale feature testing beyond the scope of our test systems. After testing, report your results by clicking the Discussion tab at the top of the page and entering:

  • Your name and email address
  • Date of your report
  • Feature test plan you are reporting on (e.g., Adaptive Timeouts, Quotas, etc.)
  • Description of the Lustre cluster/system you tested on
  • Results of your testing

Thanks for your contributions and support.

Release Testing

Test plans are available for upcoming Lustre releases.

Lustre 1.8 Test Plan

This is the Lustre 1.8 Test Plan. The objective is to find and resolve defects in the Lustre code during release testing cycles, with an ultimate goal of delivering production-ready Lustre code.

Lustre 2.0 Test Plan (coming soon)

Feature Testing

This section lists the feature test plans used by the Lustre QE team.

Adaptive Timeouts

This is the Large-Scale Test Plan for Adaptive Timeouts. The goal is to test the Adaptive Timeouts feature on a large-scale system.

CMD

This is the Large-Scale Test Plan for CMD. The objective is to test the Clustered MetaData (CMD) feature at scale, and verify that performance improvements are seen with CMD, as compared to a single MDT configuration.

NOTE: CMD is designed for use only on non-production Lustre systems.

OSS Read Cache

This is the Large-Scale Test Plan for OSS Read Cache. The goal is to test OSS Read Cache on a large-scale system to be sure the feature scales, and verify that performance data, when OSS read cache is enabled, is better than data when the feature is disabled.

OST Pools

This is the Large-Scale Test Plan for OST Pools. The objective is to test OST Pools on a large-scale cluster to make sure the feature is scalable, and verify that file creation performance is the same (with or without OST Pools).

Quotas

This is the Test Plan for Lustre Quotas. The test plan's objective is to:

  • Thoroughly test the quotas feature (block and inode quotas for users and groups) for each new Lustre release (during pre-release testing)
  • Focus on quotas hard limits
  • Find and document problems with quotas (and add them as release blockers).
  • Deliver working quotas feature code to Cray, Inc. with each new Lustre release and avoid bugs (especially regressions)

The components to be tested are:

  • ldiskfs
  • Lustre
  • lustre quotas feature code

Lustre Security

This is the Test Plan for Security Features. The test plan's objective is to find defects in the Lustre 2.0 security features and drive the security source code to a stable, production-ready state. The test plan contains:

  • Complete acceptance-small (acc-sm) run that shifts among different security flavors (null, krb5i and krb5p) between tests.
  • Complete I/O benchmark (IOR and mdsrate), to run against all Lustre-supported security flavors (null, plain, krb5n, krb5a, krb5i and krb5p), gather and analyze performance data.

Version-Based Recovery

This is the Large-Scale Test Plan for Version-Based Recovery. The test plan's objective is to verify that the VBR feature works with a large system.

Regression Testing

This section lists regression test plans used by the Lustre QE team.

Lustre 1.8 Performance Regression Test Plan

This is the Lustre 1.8 Performance Regression Test Plan. The goal is to test each Lustre 1.8 feature for performance regression on an at-scale cluster.