HLRN NHR provides tailored WORK filesystem file systems for improved IO throughput for of IO intense job workloads.

...

Default Lustre (WORK)

This WORK is the default shared Filesystem file system for all jobs and can be accessed using the $WORK variable. WORK consists WORK is accessible for all users and consists of 8 Metadata Targets (MDT's) with NVMe SSDs and 28 Object Storage Targets (OST's) on Lise and 96 OST's on Emmy. Both using rotational using classical hard drives.

Access: $WORK

Size: 8 PiB quoted

Special File System Types

Lustre with striping (WORK)

Some workloads will benefit of striping. Files will be split transparently between a number of 2 up to 28 OSTs on Lise and up to 96 OST's on Emmy.

Especially large shared file IO patterns will perform good using stripingbenefit from striping. Up to 28 OSTs on Lise can be used, recommended are up to 8 OSTs for Lise. We have preconfigured a progressive file layout (PFL), which sets an automatic striping based on the file size.

Access: create a directory with striping using "new directory in $WORK and set lfs setstripe -c <stripsize> <dir>"

Size: 8 PiB like $WORKWORK

Local SSDs

Some Compute Nodes have are installed with local SSD storage. Files on local SSDs These node share the following properties.

• 2 TB SSD locally attached to the node
• Data on SSD will be deleted after the job is finished.
• Data on SSD can not be shared across nodes

...

• .

For single node unshared local IO this is the best performing filesystem file system to use.

FastIO

...

WORK is extended with 4 additional OST's using NVMe SSDs integrated into WORK.

Access: ask support@hlrn.de for access

Size: 55 TiB - quoted

IME - Emmy only

DDN Infinite Memory Engine (IME) based Burst Buffers and file system Cache - a fast data tier between compute nodes and the Lustre file system /scratch - is used as an I/O accelerator for I/O bound problems. This helps avoid overload on the system when a program tries to write large amount of data within a short period of time to the global parallel file system. IME servers consist of Solid State Disk (SSD) that acts as a cache and burst buffer to improve the global file system performance. IME servers are currently available for use in EMMY.

random io?!

Access: IME Burst Buffer, File System Cache

Size: 48 TiB

...

to accelerate heavy (random) IO-demands. To accelerate specific IO-demands further striping for up to these 4 OSTs is available.

Access:

create a new directory in $WORK and set lfs setstripe -p flash <dir>

Size:

55 TiB - quoted

Finding the right File System

If your jobs have a significant IO part we recommend asking your consultant via support@hlrnsupport@nhr.zib.de to recommend the right filesystem file system for you.

Local IO

If you have a significant amount of node-local IO which is not needed to be accessed after job end and will be smaller than 2 TB on Lise and 400 GB on Emmy we recommend using $LOCAL_TMPDIR. Depending on your IO pattern this may accelerate IO to up to 100%.

...

Global IO is defined as shared IO which will be able to be accessed from multiple nodes at the same time and will be persistent after job end.

Especially random IO on small files will be accelerated up to 200% using FastIO on Lise or IME on Emmy

INTERNAL

Recommendation Matrix:

Max performance gain on IO versus default $WORK in brackets.

...

.

...

FastIO stripe=4       (+200%) or

$WORK stripe=4-8 (+200%)

...

FastIO stripe=4       (+80%)

$WORK stripe=4-8  (+70%)

...

FastIO stripe=4         (+120%)

$WORK stripe=MAX (+90%)

...

FastIO stripe=4         (+200%) or

$WORK stripe=MAX (+150%)

...

FastIO stripe=4         (+100%)

FastIO                        (+50%)

Vorschlag Ankündigung

...