Hawk installation schedule

This page will be updated as new information becomes available. Please be aware, HPC systems are made out of leading edge components. If one of this components is delayed, the complete schedule will change. Do not take this schedule too serious...

time frame	action
end of May 2019	Installation of a small test system
September 2nd ... 6th	maintenance and preparation of infrastructure (1st part).
November 19th	Reduce Cray XC40 HazelHen to ~4488 Nodes and preparation of infrastructure for Hawk
January 17th 2020	Installation of 8 racks Hawk, storage, pre-post servers, ... integration, testing, setup for users
until Februar 18th	complete Hawk hardware installation
mid-February	Test Phase for pilot users ~ 3 weeks. Due to the delayed delivery, the test phase will also be postponed. The filesystem will be ws9 (Hazel Hen)
February 24 2020	Final shutdown and decommission of Hazel Hen and preparation of infrastructure for the complete installation of Hawk.
February 25 2020	hazelhen workspace filesystem (ws9) has been fully integrated with all workspaces into the hawk system.
until March 1st	testing, and integrate additional racks into first phase (2048 nodes)
March 9th	general availability for all users
until March 15	prepare power and cooling facility for second phase hawk
July 15th	connect racks 1 ... 16 with tested racks 17 ... 40 This step may need 15 days. We will try to run user jobs on racks 41 ... 44
August	System functionality test. (Large applications, IO-performance, MPI - performance / funtionality, power consumption, system stability, ...)
end of August	System Acceptance. Start of production
September	integrate racks 41 ... 44 into production system
August	test and integrate all remaining cabinets into the hawk system (5632 nodes total)
even later	Acceptance and production
August	Data migration by users from ws9 (Hazel Hen) onto ws10 (Hawk) filesystem

Terms of Use

we are pleased to provide early user access to Hawk.

Please note that the system is still far from production status in terms of stability / performance / configuration / usage.

• Both the node configuration (such as numa domains per socket) and the InfiniBand configuration are not yet final und both will be subject to change.
• This means that the performance of the system is not optimal. It also implies that the users should not yet attempt an optimization of their applications based on the current setup.
• No monitoring system active, this may cause failed jobs if compute nodes break down and will be part of subsequent jobs

The usage is granted under the following conditions:

• Do not publish performance measurements of the current system configuration
• Do not monopolize the system. Give other users the opportunity to use the system. Avoid long running jobs.
• If you encounter problems, please report this to the prepared Trouble Ticket System via email to:

   rt-platform-hawk@hlrs.de

• Due to the current state of the system, your application may not yet be operational. In this case, please wait a few more days. The system will be improved very quickly.