Sunderland University (UK), Grid Computer
On the 9th Feb 2009 I went along to an IET evening lecture hosted by the University of Sunderland Centre for Internet Technologies. It was about the grid computer they have built there where you can read more about it on their site.
There is also a quick item about the event here (I’m at the front),
IET presentation on 3D computer rendering using the grid computing facility at the University of Sunderland.
CIT has trialled four different rendering applications on the grid - Autodesk 3DS Max, Maxon Cinema 4D, NetTek Lightwave and Blender - and the presentation focused on the outcome of the trials.
The lecture was well attended by a mixture of IET members, students, and others with an interest in the subject. Following the presentation, a practical demonstration of the grid was given to attendees.
I have to say I was really impressed at the environment the students there get to work and the grid computer facility itself was interesting. Points that stick in my mind were;
- Forty nodes in the grid.
- The grid computer lives in an open plan terrace with no specific air conditioning, thus the individual servers are given lots of space each in the racks (dell hardware).
- It is unique in that it dual boots into Scientific Linux and/or Windows 2003 Server Compute Cluster Edition. They are upgrading to windows 2008 HPC (High Performance Computing) edition shortly. Microsoft have taken an interest in what they are up to.
- They have a fantastic Cisco switch (6500 series) linking the grid together the view was that the linking was worth investing in thus less nodes were bought, instead a more pricy interconnect backbone.
- The grid has been used amongst other things to produce 3D renders using the above quoted packages.
- Some nodes can run linux while others run windows server at the same time.
Grid computing and 3D rendering
It seems to have taken a while to get these 3D render packages working in the grid environment. Problems with bottle necks at the network and the lack of high end graphics card hardware (GPU) on the servers. Something that is only really available in workstations.
There have been problems getting the applications to play with the architecture, mapping drives, using Network Attached Storage etc.
There were also issues with the way that the applications utilised the available nodes, often very poorly scheduling the work load of the nodes. For example not immediately giving more work to nodes that had finished ahead of others.
Research and academic world vs business world
One striking thing that that I found hard to change my mindset over is the way research academic world is so different to the business world I operate in. Here are these academics with great intentions and some wonderful kit, but learning on the job without the what I felt was appropriate support. With the size of this investment should they not have employed a good computer engineer to support the work?A good all rounder in routers, network configuration and data centre management. Someone who understands the power of virtualisation (ESX) and technologies like hypervisor.
It felt like the academics had used a lot of time learning about computer networking and how to run a data centre (for example how much freedom to give users) rather than concentrating on the bigger academic questions this system was built to answer. This is a classic example of not getting return on the investment because the adequate skill and knowledge support was not in place from the start. I’m not critical of the individuals because I am impressed by what they have managed to get up and running. I suspect technician support is one of those areas that has been eroded all over the academic world over the years. I felt that there should be a computer engineer on a similar salary to the lecturers working together to output some great research here. Is this the advantage bigger research companies like HP have over the universities?
If this were mainstream business world that hardware would be working hard all day and night every day? I get the feeling it is not in reality utilised that much.Feels so wrong, I’d love to get my hands on it. Perhaps it is time for me to do a MSc in grid computing so I can get my hands dirty! :)
There was talk of business plans to offer processing services but I was not convinced that it would work. The academic and business world have different expectations, pressures, needs it is very difficult to marry the two and indeed may be counterproductive in distracting from good pure research.
Good evening thank you – got me thinking and having all this talk bout pixel shaders and rendering has now made me think about barcode recognition and what normal processing can be moved from the CPU in computers to the GPU, even when it is not graphical in nature. A further post on this may follow shortly!