Iridis News
System Status
News Room
8/02/2010 - Iridis 3 Launch Event
Supercomputer Launch Event took place on February 8th. This event marked the beginning of operations of new HPC facility Iridis 3. Speakers at the launch event included invited guests from IBM, OCF and Southampton academics. Deputy Vice-Chancellor Professor Phil Nelson chaired the event. Talks showed wide range of applications for the supercomputer in different research areas ranging from Computational Engineering, Chemistry. Genetics to Weather Predictions and Archaeology.
1/12/2009 - Iridis 3 pilot service starts
Access to the Iridis 3 pilot service is now available to all former Iridis 2 users. An email with details of how to find out more about the service has been sent out via the Iridis-announce email list. By next week we expect to extend the service to all new users who have requested an Iridis account in the last few months.
24/11/2009 - Iridis 3 pilot service news
We had a few problems with the GPFS file store stability last week. These were traced to a dodgy infiniband switch, but this problem has been dealt with, and we are making steady progress with a pilot service that we anticipate can be extended to all former Iridis 2 users within the next week.
We've only had full access to the system with our own build for just over a fortnight, so there are a lot of loose ends to be tidied up still. Hence we have started with a handful of users, representing various research or application areas, so that we can sort out any glitches before we open up access further. This is going well so far and we are extending the number of trial users to cover more application areas (don't call us, we'll call you if we need your help!). We will send out an email via the Iridis-announce email list when the general access pilot service is available. Apologies for the absence of HPC facilities over the last couple of months but the wait for Iridis 3 should soon be over.
18/11/2009 - Iridis 3 rated 74th in worldwide top 100
Iridis was ranked 74th in the world this week in the Top 500 Supercomputer list, and is by far the fastest University-owned supercomputer in England.
13/11/2009 - Iridis 3 initial acceptance test
Iridis 3 completed the first (and major) stage of its acceptance tests earlier this week and we have finally been able to get full access to the system without being constrained by work being carried out by the contractors and related parties. We have now pushed out our initial build over the entire system and have corrected a few initial glitches. We already have candidates for a hand-full of test users who will help us check that there are no major issues that need to be addressed before we put a large number of users onto the system. These users will be contacted shortly (don't call us, we'll call you ...) and we should be able to give them access by early next week. If there are no major problems revealed at this stage, we expect to be able to open up a full pilot service to all former Iridis 2 users a few days later. We will publish an update towards the end of next week and will publicise the start of general access to the pilot service via the Iridis-announce email list.
08/10/2009 - Iridis 3 progress
Progress has been steady, initial minor problems with the Inifinband interconnect have been diagnosed and connected, and all other hardware has been checked out, with only a few minor problems still to be addressed. The O/S build is coming on well, and the basics of the system software and scheduling setup are in place. We are now in a position to concentrate on the outstanding application software builds and prepare for the benchmarking and system acceptance tests. The latter are likely to take a couple of weeks to complete, but we expect to be able to offer an initial pilot service soon after, probably towards the end of October.
02/10/2009 - Iridis 3 has landed
All the Iridis 3 hardware was delivered on Saturday 26 September. Work on installing the final connections for the water-cooling started on the Sunday morning, with power connected up on the Monday. By the end of the week the water-cooling was fully tested and commissioned allowing us to power up all racks at once for the first time, so that whole system testing could begin.
23/09/2009 - Iridis 3 arrival - Decommissioning of Iridis 2
We now have confirmation from IBM that all the Iridis 3 racks will be arriving on the 26th September. Please note that the Iridis 2 service will shutdown for the final time on Friday 25th September at 08:00, as all the space it occupies is required for the Iridis 3 installation. However, the file systems (home and scratch) from Iridis 2 will remain available and accessible via the main login node, during the transition period, once we have moved the fileservers to a new location.
The work to set up the new system, and to integrate it in to our infrastructure, will start immediately. However, the testing of the cooling, all the system components and the overall system performance will take quite a while to complete. While acceptance tests are in progress, we are hoping that we can offer a limited pilot service, on an intermittent basis, by mid-October. At this stage there will be significant interruptions at times, when tests need to be run on the entire system. By the beginning of November we should be in a position to offer a more consistent pilot service on the full system. This will lead into a full production service in the new year as we refine the configuration to get the most out of the new system.
If lack of access to HPC facilities during October is likely to cause you extreme difficulties please let us know, though our ability to provide access to alternative computational facilities will be very limited. We apologise for the inconvenience caused during the transition, but anticipate that this will be more than justified by the vastly increased power of the new system.
4/09/2009 - Decommissioning of Iridis 2 -update
It is now looking as if will not receive all the Iridis 3 racks next weekend, and therefore we will not need to decommission the Iridis 2 racks for another week or two, until nearer the end of September. We will confirm when we hear more from IBM.
4/09/2009 - Iridis 3 user environment
While the user environment on Iridis 3 will will be similar to that on Iridis 2, there will be some significant differences and a few details that might catch you out. See the page on Diffferences between Iridis 2 & 3 for a first attempt at summarising what we think the main differences will be in advance of the system arriving. Some hints are provided on how you could prepare yourself for these differences in advance of the initial user service being available.
27/08/2009 - Decommissioning of Iridis 2
The hardware for Iridis 3 is currently expected to arrive on campus on September 13. As we will now receive all of the racks in a single delivery (rather than than the two phases originally anticipated), we will need to decommission most, if not almost all, of Iridis 2 before this date to make space for the 15 new racks and to allow them to be powered up so that we can test the water cooling system. We will endeavour to continue to provide access to the Iridis 2 file systems through the login nodes, but we will probably need to decommission the compute nodes by September 11th. This will be confirmed in the near future. There will then be a period until around early October, when we will not be able to offer access to the new compute nodes while we are running tests on the water-cooling, the system itself and running acceptance tests. We apologise for the lack of access to compute intensive resources during this transition period, but the logistics of the installation leave us with little choice. Further updates will be published on this page as the schedule of change firms up.
12/08/2009 - Partial decommissioning of Iridis 2
In preparation for the plumbing works required for Iridis 3, most of the single-core nodes were decommissioned at the beginning of August and the remaining 59 single-core nodes plus the Myrinet nodes and some of the dual-core nodes have been successfully moved to another location. The scheduler has been modified to allow jobs, which would have originally been targeted at the single-core nodes only, to run on the dual-core nodes. While this is not optimal it is not a major issue for the remaining life of the machine.
14/07/2009 - From Iridis 2 to Iridis 3 in a few simple moves ...
The university has just issued a press release on the contract to provide a successor to the current Iridis cluster. So at last we can tell you a bit more about the dramatic increase in computational power that will become available for researchers at Southampton (and also what will need to be done to make way for the new system).
The contract was awarded to our existing suppliers, OCF, in partnership with IBM, who tended an outstanding bid after an exhaustive procurement process in the face of strong competition.
The new system will consist of 1008 compute nodes. Each node will have 2 Intel Nehalem quad-core processors, giving 8 cores per node and 8064 processor-cores total - so roughly 8x our current core count. The main features of the new system are as follows:
- Three 8-core login nodes, each with 32 GB of memory
- 1008 8-core compute nodes (Intel Nehalem 2.26 GHz)
- Most nodes have 24GB of memory, 32 have 48 GB
- All nodes will be connected by a fast infiniband network
- 100 TB of storage supporting a high-performance GPFS file system
- Additional nodes for Visualisation and for GPU computing
- Further provision for very high memory needs
- The main system will be Linux based, but there will also be a significant number of nodes running a service using Windows HPC 2008.
The compute nodes will be housed in IBM iDataPlex racks, with water-cooled rack doors removing the dependency on the air-con units in the data-centre. Further details of the system will be published here at a latter date.
While this is very good news for the future of research computing at the university, there will be, as you might expect, disruption to our HPC services during the transition from Iridis 2 to Iridis 3. The size of the new system means that it will require all of the floor space and power (and then some ...) currently occupied by Iridis 2 and its sub-clusters. We have been putting a great deal of effort into a transition plan that will not leave users without access to HPC facilities for the entire period of the changeover. We plan to make an initial installation of 2 or 3 iDataPlex racks which should give processing power comparable with the current Iridis 2. Once we have a pilot service running on this system, users will need to migrate onto the new system immediately so that we can decommission the rest of Iridis 2 and install the remaining 10-11 racks.
While we will do all we can to minimise disruption to our HPC services during the transition, the phrase "no pain, no gain" is applicable here, and users should be prepared for significant capacity reductions, and for some periods of downtime at crucial stages during the commissioning of the new system. User-support response times on the current system may also be slower than we would like during this period of intensive activity, though we will do out best to ensure that critical issues are dealt with.
A rough timetable of events will be confirmed as soon as we have confirmation from IBM/OCF of delivery dates.
14/07/2008 - Migration of Iridis file store to new hardware
On the morning of July 14 2008 the old Iridis "scratch" file systems will be migrated to new hardware. Iridis and related systems will not be available during the transition. The "old scratch" file systems are /scratch & /scratch1 through to /scratch5. Users of the these systems should check important infomation regarding the change.
19/11/2007 - Change of default login node and migration of Iridis file store to new hardware
On the morning of November 19 2007 the Iridis file store will be migrated to new hardware and the default login node will be changed. Iridis and related systems will not be available during the transition
9/10/2007 - Node memory upgrades
A number of nodes have been upgraded from the standard 2GB of memory recently:
There are now a total of 25 single-core and 11 dual-core nodes with 4GB plus 5 single-core and 2 dual-core nodes with 8GB.
We are also trialling a dual-core node with 16 GB
Access to the 8GB nodes is currently restricted, let us know if you need to be added to the list.
31/8/2006 - Iridis web page updates
The Iridis web pages have been republished at this new location as the old iSolutions website will be shutdown latter today. Some minor revisions have been made to the content, but the scale of the changes has been limited by the need to ensure that the basic information is still available. A more comprehensive revision of content will occur over the next few months.
31/8/2006 - Iridis still growing
Most of the final upgrade to Iridis under the SRIF 2 funding is now in active service. The latest upgrade includes:
- 82 New dual-processor, dual-core nodes for general use - giving an extra 368 CPU cores.
- A fast Myrinet network on 64 of the older single-core nodes.
- Extra memory on some nodes - 14 nodes with 4 GB and 2 more with 8 GB to add to the 3 existing 8 GB compute nodes.
- Extra nodes funded by individual groups and primarily dedicated for their use.
Iridis 2 now has well over 1000 CPU-cores. See the latest developments page on how to use the dual-core nodes and the nodes on the Myrinet network.
News feeds