Details
-
Improvement
-
Resolution: Unresolved
-
Minor
-
None
-
None
-
3
-
9223372036854775807
Description
The following pages are all over 3 years old. These are being reviewed as part of the efforts of the Lustre Documentation Working Group. The main intent of this bug is to evaluate if these pages require:
- Reorganization
- Removal due to staleness
- Extension due to missing material
- Correction to fix incorrect or stale material
The author is expected to annotate this work item with the changes expected. They may or may not make those changes as part of this work item. If they choose to not make those changes, they are expected to create a new work item where the identified, explicit improvements can be made by another individual within the LUDOC stream. If they do choose to make those changes, please clarify such in the comment section.
The full list of pages include:
Performance and Tuning:
Large-Scale Tuning for Cray XT (07:43, 25 August 2016)
Tuning Lustre for Performance (07:19, 13 October 2016)
SGPDD Survey (23:14, 25 June 2017)
OBDFilter Survey (00:25, 26 June 2017)
LNET Selftest (19:01, 19 October 2017)
VDBench (04:01, 29 June 2017)
Optimizing Performance of SSDs and Advanced Format Drives (22:53, 30 August 2017)
MDTest (10:32, 22 June 2018)
IOR (22:31, 16 May 2019)
Optimizing o2iblnd Performance (09:23, 29 July 2019)
Lustre Tuning (18:26, 2 October 2019)
Attachments
Activity
LNET Selftest – Still up-to-date minus a broken link to the manual (fixed) and /bin/sh needed to be revised to /bin/bash as 'let' isn't a valid command in plain sh (fixed). Tested the script afterwards and it functions appropriately on my cluster.
OBDFilter Survey: Still up-to-date minus a broken link to the Lustre manual when it used to be hosted by Intel. Fixed, and retested obdfilter on my cluster and it still works.
SGPDD Survey: This benchmark seems to still be valid. It is present on my Lustre cluster by default, as is the sgp-utils package it relies upon. Attempts to run the example commands on my system result in:
starting io on sg device, error: Invalid argument
sgp_dd: inputting to sg failed, blk=1024
I haven't root caused this, but it's entirely possible it's because it's attempting to run with a default block size that's not amenable to the Azure managed disks I'm pointing it at. I'm putting a pin in this one however as broadly this page looks up-to-date and still useful if you want a survey-like script to evaluate your LUN/RAID before system bring-up.
I updated broken links on the page to point to the correct location within the Lustre Manual.
Tuning Lustre for Performance: This page had six broken links, and as it is redundant in nature to "Lustre Tuning," which I found more aptly named, I have deleted it. No pages linked to it.
Large-Scale Tuning for Cray XT: This page provided no source for how the authors arrived at this tuning, and referred to a system (cray xt3) that launched 20 years ago. I have deleted it, and removed now-broken links from the following two pages:
https://wiki.lustre.org/Tuning_Lustre_for_Performance
https://wiki.lustre.org/GetInvolved:Get_Involved
Optimizing Performance of SSDs and Advanced Format Drives – this is a straight C&P from the ZFS FAQ. Rather than plagiarize and needlessly duplicate, I removed reference to this page in https://wiki.lustre.org/Category:Lustre_Systems_Administration, and have instead included a brief section that links to the FAQ on https://wiki.lustre.org/ZFS_OSD_Hardware_Considerations.
VDBench - Downloads link from Oracle was out of date, so updated it. Fixed a few grammatical errors, and removed references to the old SF version of it. Otherwise the directions look roughly correct, and VDBench has only moved 1 patch version since these were written (it isn't actively maintained). We still some folks using this tool, so I'm leaving this page around in case people still want to play with it.