Stéphane Thiell

I do HPC storage at Stanford and always monitor channel 16 ⛵

[Stéphane Thiell @s_thiell]

The agenda for the Lustre User Group 2024 is now available. I hope to see you in Lubbock, Texas! https://t.co/tVwewIgRmr

[Stéphane Thiell @s_thiell]

Our @StanfordCompute branch of robinhood-lustre-3.1.7 on Rocky Linux 9.3 with our own branch of Lustre 2.15.4 and MariaDB 10.11 can ingest more than 35K Lustre changelogs/sec. Those gauges seem appropriate for Pi Day, no?
https://t.co/DewirkKhml
https://t.co/PPHsPJPFam

[Stanford Research Computing @StanfordCompute]

We are hiring! Stanford Research Computing is recruiting a #GPU cluster sysadmin, for an incoming @nvidia #DGX SuperPOD.
If you're interested in operating large-scale GPU systems to support the amazing @Stanford research community, feel free to apply at
https://t.co/zR1KmJ5bjX

[Stanford Research Computing @StanfordCompute]

#Sherlock goes full flash!
The cluster scratch file system has been revamped to provide 10 PB of fast flash storage on #Lustre
https://t.co/oXVWoRsZRP
#HPC #Stanford

[Stéphane Thiell @s_thiell]

@BPCZed @StanfordCompute Yep! Not a system designed for hero IO-500 benchmarks. Our users needed the most a scratch space for very large data sets with random I/O capability with a long retention period (we purge at 90 days on data_version). Cost was kept under control thanks to high capacity SAS SSDs.

[Stéphane Thiell @s_thiell]

Filesystem upgrade complete! @StanfordCompute cares about HPC I/O! The Sherlock cluster has now ~10PB of full flash Lustre scratch storage at 400 GB/s, to support a wide range of research jobs on large datasets! Fully built in-house!

[Stéphane Thiell @s_thiell]

@aryaman2020 @saizhang0 Sherlock?

[Stéphane Thiell @s_thiell]

Congrats to the authors of the paper “Xfast: Extreme File Attribute Stat Acceleration for Lustre” presented at #SC23 covering 10+ years of improving stat operations and exciting new features coming soon! 👏

[Stéphane Thiell @s_thiell]

Intro of the new xCAT consortium at #SC23 with 60+ attendees (including remote ones) and IBM on the call to facilitate the transition. Room booked by Dell (Omnia is using xCAT). There is hope.

[Corey Quinn @QuinnyPig]

Allow me to be more helpful to @SlackHQ customers than their entire support apparatus:

If you’re using the native Mac app, here’s how you disable their dogshit user-hostile redesign:

[Stéphane Thiell @s_thiell]

@glennklockwood TS1170 is not faster than LTO-9 (both at 400 MB/s) which is also interesting.

[Stéphane Thiell @s_thiell]

Today we reached a total of 4.9 PB of flash storage at 200 GB/s for @StanfordCompute’s Sherlock to support a wide range of research workloads. Lustre 2.15. #HPC #AI

[Stéphane Thiell @s_thiell]

Temperature vs Drive Size, from ⁦@backblaze⁩ #MSST - the introduction of Helium originally helped but the trend is up again!

[Stéphane Thiell @s_thiell]

⁦MSST’23 @SantaClaraUniv⁩ has started. Great first day with Trond Mykelbust, the maintainer of the Linux kernel NFS client, presenting NFS 4.2 improvements (including pNFS!)

[A. Karl Kornel supports 🇺🇦 @californiaKARL]

Are you a Facilities-type person? Like electrical/HVAC? Work (or wanna work) in a data center? I've got a job for you… https://t.co/GRF3iiRG8b

[Stéphane Thiell @s_thiell]

@b4mb0u @OpenSFS @Whamcloud @awscloud @ddn_limitless @ORNL @TACC @Livermore_Comp @Azure @HPE @argonne @UCSanDiego Sadly no Eviden talks! Too busy rebranding? But registration is still open... :)

[Stéphane Thiell @s_thiell]

The agenda for the Lustre User Group 2023 (LUG'23) is available at https://t.co/Y3PGzdSmJI
featuring talks from @OpenSFS @Whamcloud @awscloud @ddn_limitless @ORNL @TACC @Livermore_Comp @Azure @HPE @argonne and more!
Hope to see y'all at @UCSanDiego in early May.

[Stéphane Thiell @s_thiell]

@aizhar @glennklockwood @StanfordCompute @Stanford Lower cost + excellent redundancy and scalability of the SAS backend. No NVMe for us at this scale yet. Maybe one day!

[Stéphane Thiell @s_thiell]

Now 2.5PB of usable flash storage at 100GB/s is in production for @StanfordCompute’s Sherlock /scratch. It’s just the beginning! High capacity SAS SSD w/ robust TLC NAND used with #Lustre over IB HDR is making a huge difference for research at @Stanford! #HPC

[Stéphane Thiell @s_thiell]

We are upgrading @StanfordCompute #HPC Sherlock's #Lustre /scratch to full flash (see my LAD'22 talk), but our old filesystem (from 2018!) is still doing pretty well for sequential I/Os. HDF5 FTW.