[Jlab-scicomp-briefs] August 19 2024 Planned Outage of farm, ifarm, and storage for Lustre Upgrade

[Bryan Hess]

Planned Outage of farm, ifarm, and storage for Lustre Upgrade
Monday, August 19th 9:00am – Wednesday, August 21st 9:00am

The farm, ifarm, SWIF, Globus, tape storage, and Lustre storage (/cache and /volatile) we be offline during this time window to transition to a new Lustre parallel filesystem. Farm nodes are reserved starting August 19th at 9:00am. Farm jobs that are scheduled to run during this period will be queued until the work has been completed.
Files stored in /cache and /volatile will be copied to the filesystem with the following caveats:

  *
If the final filesystem synchronization runs long, we will stop it at 9:00am on August 21. Any uncopied files will be limited to those written in the days just before the upgrade. We will maintain /lustre19 as an offline backup for at least one month.
  *
The /lustre19 path will not be valid after the change. Please use /cache and /volatile in absolute paths to files.
  *
As a reminder, we are scheduling an outage for this transition because of a limitation in older hardware that prevents us from running both Lustre filesystems simultaneously.


Steps to help facilitate a speedy Lustre transition on August 19th:

  *
Consolidate or delete small files on Lustre if possible
  *
limit directly writing large file sets to /cache or /volatile the weekend before the upgrade. This will help to keep the final file system synchronization from running long. Important Note: Using jcache is not a concern because jcache requests are written to both the old and new Lustre filesystems.


A knowledge base article with configuration information for hosts that access Lustre over NFS will be published soon.
The new Lustre filesystem increases storage for /cache and /volatile from 4.7PB to 11.2PB and increases overall throughput.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/jlab-scicomp-briefs/attachments/20240809/2de494b9/attachment.html>