Sherlock changelog

GPU compute modes

by Sherlock Team
We developed a new extension to allow users to change the compute mode of the GPUs allocated to their jobs. By default, GPUs are set in the "Exclusive Process" mode for best performance and ensure that only one process can use the GPU...
Improvement
New

New login nodes

by Sherlock Team
Two new login nodes, sh-ln03 and sh-ln04 are now available to connect to Sherlock 2.0. They've been added to the load-balanced DNS pool and will automatically be selected according to their load and the number of currently logged-in...
Improvement
Hardware
New

RSS feed

by Sherlock Team
To make it even easier to follow news and updates, the Sherlock Changelog now supports RSS feeds! Just add the following URL to your favorite RSS reader: > https://news.sherlock.stanford.edu/feeds and be sure to never miss an update...
Improvement
Announce

Cron jobs on Sherlock

by Sherlock Team
Cron jobs are not supported on Sherlock, and going forward they won't be allowed anymore. Due to various past incidents, we had to disable the functionality, to prevent runaway cron jobs to take out entire login nodes. Moreover, compute...
Improvement

New purge policy on $SCRATCH

by Sherlock Team
After holding up for a long time, we're implementing a new expiration policy for all files stored in $SCRATCH and $PI_SCRATCH on Sherlock. Starting October 1st 2017, all files in $SCRATCH and $PISCRATCH that have not been modified...
New
Announce

New $SHERLOCK environment variable

by Sherlock Team
As users start trying out their job submission and application scripts on Sherlock 2.0, we figured it would be useful to easily determine on which platform those scripts are running. So we've added a new $SHERLOCK environment variable...
Improvement
New

GPUs now available on Sherlock 2.0

by Sherlock Team
The new gpu partition has just been open for production on Sherlock 2.0. It currently features two kinds of accelerators: NVIDIA Tesla P100-PCIe GPUs, for applications requiring double-precision (64-bit) and deep-learning training...
New
Hardware

New Lustre routers in production

by Sherlock Team
Two new Lustre routers have been put in production on Sherlock 2.0, one to serve /scratch, and the other one to serve /oak. Immediate benefits will be higher availability of the filesystems on Sherlock 2.0 clients, and doubled bandwidth...
Improvement
Hardware

Sherlock 2.0 is here!

by Sherlock Team
We're very glad to announce that the long wait is finally over: Sherlock 2.0 is now available for testing! Currently available to a restricted group of users for beta-testing, the new cluster is expected to enter full production in the...
New
Announce