How has it helped my organization?
We have been working with the Pittsburgh Supercomputing Center for around ten years. They are picking the hardware and they had picked this hybrid system. It has several different kinds of components in the system and we had worked with them for a long time. We knew that they were picking the stake of that stuff so that's why we selected this solution.
What is most valuable?
It's very hard for a professor to amass the supercomputing resources, so I've been very fortunate to have that level of supercomputing at our disposal and that has really enabled us to do the world's leading superhuman AI research. That is what we did, we actually beat the best heads up in all Texas, holding human players in the world this January. So, we're at a superhuman level in the strategic reasoning.
What needs improvement?
One thing that we are looking for is the better stability of the Lustre file system, it could be improved. I have heard that they are coming out with a better memory bandwidth, so that's good or maybe, it's already there in System 10.
In that case, of course, then there is need for more CPUs, more storage and all of that.
What do I think about the stability of the solution?
It has been fairly reliable. In the beginning, of course not, but then we were a “baiter customer”, so in the beginning, there was nothing, literally there was nothing in the racks. We've been with it from the beginning and of course, in the beginning, it was less stable. However, it became more stable over time.
If there's anything that hasn't been that stable, then it is the Lustre file system. I would say that they have made some improvements with that but this is not just a problem with bridges. We have computed the other supercomputing centers like San Diego Supercomputing Center in the past as well and Lustre seems to be just a little bit unstable overall.
What do I think about the scalability of the solution?
It's going to meet our needs moving forward, it is scalable. Having said that, our algorithms are very compute-hungry and storage-hungry, so more is more and there's no limit as to how much our algorithms can use. The more compute and the more storage they have, the better they will perform.
How is customer service and technical support?
I would support the Pittsburgh Supercomputing Center (PSC) support; they gave us the support and their support has been awesome. We don't directly contact HPE, they contact HPE if needed.
How was the initial setup?
The PSC installed everything, i.e., both hardware and software. So we didn't do any of that; from our perspective, it has been easy to use.
What other advice do I have?
Whilst looking for a vendor, we do not look at the brand name at all. Instead what we look for are just reliability and raw horsepower.
It has been great. The Pittsburgh Supercomputing Center guys have been great in supporting us very quickly and sometimes even at night or on weekends. I've been very fortunate as a professor to get this level of supercomputing, so we've been able to do the world's leading research in this area. The only things that I would improve are the ones that I have mentioned before, i.e., the Lustre file system, and maybe, the memory access from the CPU.