|
|
||||
|
Bulletin 195 - 2009 November 30
1. CSIRO SX-6/TX7 system access We plan to turn off nodes sx600 to sx604 (the CSIRO nodes) on 1st December. Access to the CSIRO filesystems through the TX7s will be available for a while after that. However, in order to release sufficient disc spares for operational needs, we anticipate turning off the discs under the /cs/flush1 and /cs/flush2 ($WORKDIR and $TMPDIR) file systems, followed by the /cs/data1 and /cs/data2 ($DATADIR) file systems. Users should start reviewing their holdings on all the SX-6/TX7 file systems. [ page top ] 2. CSIRO File access times, and flushing IMPORTANT NOTICE TO ALL USERS OF CHERAX AND BURNET The /cs/flush and /work areas on burnet and cherax are flushed to free space when they become too full. The flushing script sorts files by the most recent of access and modify times, and removes files from the oldest to newest until a threshold is reached. Recent feedback indicates that the flushing was not always working correctly. Subsequent investigation found that access times were not always being updated when a file was accessed on burnet and cherax, related to evolving XFS support in the linux kernel. Note that you can see a file's access time with ls -alu or stat. and with this bug the access time listed for a file is updated when accessed, but may revert to the old time (presumably the access time recorded on disc) after a short interval. We did not know about this feature. Although our flushing algorithm appears correct, it relies on the access times and it is not updated properly, files recently accessed may still be considered candidates for flushing. We have installed a fix for the problem on cherax on 25/11/2009; however, we have not located a fix for burnet. Until the fix is installed, we recommend:
Note that the TX7s appear not to exhibit the problem. DMF on cherax is likely to also have been adversely affected by the problem and may now get better performance with better retention of active files on the main disk. Lustre, as in use on the NCI and Bureau Sun Constellation systems, has the following characteristic ( from http://wiki.lustre.org/index.php?title=Lustre_FAQ ) "It is not practical to maintain fully coherent atime updates in a high-performance cluster file system. Lustre will update the atime of files lazily -- if an inode needs to be changed on disk anyways, we will piggy-back an atime update if needed -- and when files are closed."on cherax is still the best place to keep important, large data files. See http://www.hpsc.csiro.au/userguides/ds/ for information about the data store. [ page top ] 3. Australian Research Collaboration Service (ARCS) ARCS is seeking recommendations on how to invest $35 million in ARCS-managed activities, providing core infrastructure and services that establish the national collaboration and data fabric. The ARCS Mission is to enable and enhance research through the provision of long-term eResearch support, services and tools. The home page is at http://www.arcs.org.au/ If you collaborate around data, or have need to access or visualize data from more than one site, then these proposed services will be of interest to you, and you are invited to comment at http://www.arcs.org.au/rfoc by 18 December 2009. Explanation of the proposed services are contained in documents attached to the ARCS Request for Open Comment form and also at http://www.arcs.org.au/rfoc/files/ including a printable brochure version of the form which you could post on your notice-board. Please take this opportunity to have your say about which of these proposals are most likely to benefit your research and/or community. [ page top ] 4. CSIRO ASC Software Upgrades MATLAB We have recently added 2 MATLAB core licenses due to the high demand. Materials Studio 5.0 Some of the highlights of this release are:
For more information, go to http://accelrys.com/products/materials-studio/ Materials Studio is available on burnet and can be used on your local workstation. Please note that in order to run remote Materials Studio jobs on burnet, users must be registered. Request registration for Materials Studio via Please see the NCI software map for usage instructions. http://nf.apac.edu.au/facilities/software/index.php?site=CSIRO under 'Computational Chemistry' [ page top ] 5. CSIRO Supplementary information on the Vayu system at NCI With the commissioning of the Vayu Sun Constellation at NCI and the impending decommissioning of the SX-6 at BOM, many of CSIRO users' jobs will move to running on the NCI. Following are some tips for getting started on the NCI systems Key information for CSIRO researchers accessing NCI is at: http://www.hpsc.csiro.au/facilities/index.shtml#nci Most information is at the NCI NF website http://nf.nci.org.au and we only aim to supplement this with tips that are specific to CSIRO (and CAWCR) researchers. You can apply for an NCI account at http://nf.nci.org.au/accounts/ Follow the link to "NEW Partner project" near the bottom of the form. 5.3 SSH and moving files between cherax and vayu NCI accounts do not have the same user name as your CSIRO accounts. You need to specify the correct remote-user name. It is straight-forward to ssh from cherax to vayu (on the normal ssh port - 22), but ssh into CSIRO from outside is generally blocked. We run an additional listening ssh service on cherax on port 22000 that you can use instead, but it only allows key-based access. The service uses hpn-ssh which is patched to provide higher performance than usual with ssh. Your home directory on cherax is still the best place to keep important, large data files. See http://www.hpsc.csiro.au/userguides/ds/ for information about the data store. The main methods for file transfer are scp and rsync, which both use ssh for the underlying connection so you need to have that working and hpn-ssh will provide faster transfer rates. If you are not using hpn-ssh, you will see transfer rates of about 7MB/s from vayu to cherax (faster from cherax to vayu). If you *are* using hpn-ssh, the transfer rate will be more like 20-30MB/s. If you are transferring many small files ("small" == less than about 10 MB) you will get the best performance by using rsync. If you are transferring a few large (100's of MB or more) you will generally get the best performance from scp. For more detail, see:
5.4 Using the NCI system and getting help See the quickstart and user guides at http://nf.nci.org.au/facilities/ for information about vayu and running jobs via the batch queuing system. For queries needing vayu system administrator support (such as requesting that a piece of software be installed), read the help pages at http://nf.nci.org.au/help.php and contact the NCI helpdesk directly at For queries related to using the NCI systems in conjunction with CSIRO ASC systems, contact the ASC helpdesk at For other queries, you can contact either and we will co-ordinate. Please tell us if you have problems or are hampered by system limitations. We can often help with advice on how to make good use of the systems and relax restrictions in cases where that is sensible. [ page top ] The CSIRO GPU cluster was successfully launched at the CSIRO Discovery Centre on Wednesday. So far the system appears to have generated significant and favourable interest from the national research community and the electronic media. A key feature of the launch was presentations by three of the initial users, describing their research applications and their initial scaling results on the cluster - some impressive, world-leading achievements. Alex Zelinsky commented in his address that the system brings CSIRO "back onto the bleeding edge, where we belong", in scientific computing technology as well as computational science applications. The neat juxtaposition with last week's 60th anniversary of the commissioning of CSIRAC was also highlighted. There was a well-deserved acknowledgment of the IM&T staff and partners who have worked hard together to get the system up and running in time to demonstrate its capabilities at the launch. See http://www.csiro.au/resources/GPU-cluster.html and/or send e-mail to for more information. 7. CSIRO GPU Acceleration of Bioscience (GAB) Grants The GPU Acceleration of Bioscience (GAB) Grants 2009 were announced at the launch of the GPU Cluster GAB Grants are open to all CSIRO staff and are intended to encourage uptake and adoption of new computational capability within the biosciences. See: http://wiki.cmis.csiro.au/wiki/index.php?title=TransBio/Call/20091125 (CSIRO internal only)
|
|
Comments to: © Copyright 2010, CSIRO Australia Use of this web site and information available from it is subject to our Legal Notice and Disclaimer and Privacy Statement |