In the UK there are some medium sized HPC systems being installed, funded by the Research Councils. The tier 2 HPC launch event on 30 March 2017 – and made announcements by the EPSRC CEO Philip Nelson and from Susan Morell (Birmingham)
Peter Vincent (Imperial) presented the PyFR higher-order polynomial CFD that works on all architectures -giving 13.7 petaflops on the USA Titan system. He described VTKm for remote visualisation that rendered the 1TB per-frame data in-situ -vis:
… this claimed and showed you could stream video sets from the 1TB data file stored in the HPC memory in this virtual wind-tunnel type simulation, without saving the data first (or importantly moving it anywhere). it used a type of Catalyst within kitware’s range of tools so could produce paraview type effects.
The results are very impressive, and similar systems are being used in other remote projects. It does require:
- knowledge of what you wish to view before you compute
- unable – or difficult – to change the visualisation results (sub-parameters)
- unknown question as to if you should (or can) store the intermediate data – would take many minutes or longer to do this.
- cost of visualisation is include – or added to – the cost of HPC so needs accounting for.
My kind of desk – pre-work within Research IT office, for a touchtable and projection display stand at the University of Manchester Regius Professorship award celebration.
Using the new Acer Predator (17 G9-792) laptop, with an inbuilt great graphics card (NVIDIA® GeForce® GTX 970M with 6GB GDDR5 Dedicated Memory), and touchscreen – bits borrowed from UoM and STFC visualisation groups along with (STFC www.CCPi.ac.uk) tomography data and the drishti-prayog volume visualisation software all appear to run extremely sweet.Moving this to the exhibition area within the NGI (National Graphine Institute) with a HDMI splitter should then be suitable for 100’s of viewers per hour …
Show setup for 150+ special visitors. This example on the screen (one of about 15), just prepared, showed work from by the new EPSRC Flagship CCPi fellowship, Daniil Kazantsev et al from RAL that using multiple volume vis transfer functions highlighted the centre and the fractal edge of slices showing dendritic behaviour (even measurable).
The laptop and small workstation were hidden and all you see is the touchtable, controlled by the operator or user on the right, and the main screen allowing multiple people to view and comment at once.
The volume visualisation was a little tricky as the raw data had been pre-segmented already so I had to create a boundary edge by interpolation and reducing the size of the complete object.
There was a talk by Regius Phil Withers after a presentation of the award from the Queen’s representative, as well as a question and answer session chaired by the President of the University of Manchester.
Not really vis but this was important in the day.
Back in the Computer Lab., at the University of Cambridge in about 1992 there were at least two filter coffee systems; both running on a slow drip-filter setup. The Rainbow Group had access to one machine in a walk down a corridor to a small kitchenette; but the network team had to travel a lot longer way to their coffee machine. A problem of lack of enthusiasm to travel meant that often they would find an empty coffee pot – or even worse stale coffee. A solution was a camera video stream that linked to the linux server and stream images to X Windows. So when new fresh coffee arrived this would be obvious from the image on everyone’s desktop.
http://64bits.co.uk/ have been rebuilding old systems and in East London they rebuild the Cambridge Coffee pot scenario – with similar equipment. There is a wikipedia page on the original: https://en.wikipedia.org/wiki/Talk%3ATrojan_Room_coffee_pot
Is video streaming – visualisation – well definitely yes as it is an interpretation. Due to low quality network speed and CPUs; the small 128×128 video stream of images had to just include the coffee level and also the fps was about one frame every few seconds – but that is a good rate and good enough resolution for “coffee pot observation”.
Touchscreens are all the fashion – but any system you deploy is expected to work smoothly and intuitively – including visualisation ones. There are some company purpose-built systems for exhibitions that are far from cheap although they may work well they can also suffer from lack of agility when recreate new content. There are also simple tablet/phone software options available for touching and manipulating volumes but can have limited graphics ability.
Over the last couple of years we have collaborated informally with partners to share knowledge on open source solutions for large scale volume visualisation (say greater than 256 x 256 x 256) on a touchscreen device.
A hardware specification service has evolved where we can assist with advice and quotes for touchscreen and workstation visualisation nodes, as well as more importantly related software installation; used for exhibitions and PE. A wiki is available at:
At Manchester, connecting through Research IT, there are now groups available to informally share equipment and expertise – and as importantly share experience in interacting with the public and other scientists – even if their research data content is often diversely different.
- MRI medical data e.g. exploring white-matter changes within the human brain: Geoff J M Parker, Hamied Haroom, Saray Parkes (ISBE)
- Multiple Materials Science analysis e.g. cracks, fibres and mechanical defects: Sarah-Jane Clelland, Phil Withers (MXIF)
- Fossil and everything ancient within the Manchester Museum: Russell Garwood, Alan Brown, Campbell Price, Roy Wogelius (ICAL)
Both being asked to be available for the upcoming Presidents Office organised ‘University Celebrations of Regius Professorship’ on 25 April 2017 in the NGI.
I was having to wait for a train at Euston Station in London- and visited the Welcome Trust that is opposite, and they had an exhibition on the story of electricity.
One neat bit of 3D visualisation was near the end of the exhibition and showed a 1950s Electrical Consumption Graph:
“This 3D graph, compiled by the planners of the Central Electricity Generating Board, represents the daily electrical energy consumed over a period of two years during the 1950s.”
Unfortunately, not very interactive – with the glass protection, although you can view it easily from many angles; but a few things were right in that they chose two years of data. This is a trick used today even when there is only one year’s data available – where you recommend copying the data – and was used by Florence Nightingale in her more radical circular statistical plots (of morbidity rates during the Crimean war). The problem comes with matching the ends up – between January with December in most calendar statistics. These should be continuous, but the human eye has difficulty in spotting smoothness etc. If you can not use circular plots and in this case with card indexes this would have been very difficult, then by repeating the data or here using two years of data, you can see the changes between December and January as clearly as any other pair of neighbouring months.
In principle the whole medical topic of radiotherapy analysis is full of visualisation. A recent sandpit discussed Monte Carlo techniques that are critical within the radiotherapy planning / beamline construction / or just radiation safety / etc…
So where is the visualisation interests as the main need is to make these processes faster and more useful. We did pitch three small projects to look at:
- Solving the 1:1:1 problem may involve visualising the results immediately after compute and not storing or transmitting this.
The 1:1:1 problem termed by radiotherapists is to model 1mm cubed, solved in 1 second with 1% accuracy for practical use.
There will be a one-day workshop where the required (back of the envelope) 250x speed up will be discussed with UK researchers and industrial participants; from the software / hardware / and remote access skills.
- Sharing data discussed ways to share both code and data: an example of data / code share that worked is on zenodo (yes have used github as well).
– Data set at: https://zenodo.org/record/16474#.WMlng_ni7Lk
– Code for reconstruction at: https://zenodo.org/record/16539#.WMlnlvni7Lk
– Even has a published tutorial: http://eprints.ma.man.ac.uk/2290/01/covered/MIMS_ep2015_26.pdfThese have official DOI tags and therefore are version controlled for both code and data so then can be used in papers by others – and have been,
- A final project is looking at fractal analysis from CT to MRI analysis and considering the problem from 3D to 2D to 1D. This will enable new planning processed to occur.
More details to come. This was organised by the Advanced Radiotherapy network https://www.advanced-radiotherapy.ac.uk/ and thank them for allowing these ideas to be pitched.
There were global other issues and “Focus Where it Matters”, was a industry quote and will reuse here; so they need not just to increase speed but also to control inaccuracy and inconsistency.
Had the great privileged to be one of the judges at the Manchester Grammar School, Science Fair Projects involving about 200 very enthusiastic future scientists. Each group, of three, had to design and evaluate a scientific hypothesis applied to sport, which was the theme for this year; e.g. do balls roll faster with change in temperature, does sugar consumption relate to concentration performance analysis, …
All did very well, but there was a good debate and points raised on how to present their work results- should they use bar charts, line graphs,and what scale ranges are important and if error outliers should be included. Interesting discussion also occurred on the quality of the graphs, produced automatically by computer software programs and how to change these.
It was good at this age that they could grasp some of the concepts that seem illusive to certain older more mature researchers when they present their work. A key lesson learnt was not to spend all your effort creating wonderful results, if the presentation and visualisation story is not given a proportional amount of time to be created.