Cray User Group: Day Four
May. 12th, 2016 05:44 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
sawylUp at dawn for the Altair open source BOF. I let Bill talk me into entering their prize draw for a drone which, to my great surprise, I won! Once the shock had worn off and cold, hard reality had set in, I realised our ethical guidelines prevented me from accepting it so after th session I apologised and discretely asked if them to try again, which they did.
The first part of the morning was dominated by a series of presentations on monitoring and resource tracking. I was particularly interested by information that Sandia's LDMS can be used to collect interconnect statistics on compute nodes with minimal impact on the running jobs.
I particularly liked the paper on pcircle FCP which described Oak Ridge's parallel copying and diffing tool. There was some interesting fundamental computer science — distributed parallel hashing and termination detection — coupled with details on the messiness of the implementation and the compromises needed to implement user-friendly features like checkpoint and progress reporting.
At lunch, CW finally explained the green dot on my badge and on CM's badge: it doesn't indicate a site that hasn't paid; instead it's a sign that the owner is a vegetarian — not that the veggie lunches were anything to write home about.
The afternoon was application focused and I made a point of attending both LA's and MG's presentations, both of which were good although I'm not sure I'm ever going to understand plasma physics. I followed this with a talk on drain times which was interesting but which relied on output from Moab to implement.
The last couple of sessions were focused on SSDs, firstly an overview of Intel's new NVMe devices, followed by a presentation by DDN on their Infinite Memory Engine — essentially a neat way of placing a burst buffers in front of standard parallel file systems to improve the performance of oddly shaped IO requests.
And with that the conference was essentially complete.
The technical sessions were excellent, I talked to a lot of people — but I'm acutely aware there were lots I missed — and it was good to catch up with people I haven't seen for ages. I've got lots of useful ideas to take home with me and it's inspired me to do more fundamental stuff over the next year and to try to put together a paper for 2017.
The first part of the morning was dominated by a series of presentations on monitoring and resource tracking. I was particularly interested by information that Sandia's LDMS can be used to collect interconnect statistics on compute nodes with minimal impact on the running jobs.
I particularly liked the paper on pcircle FCP which described Oak Ridge's parallel copying and diffing tool. There was some interesting fundamental computer science — distributed parallel hashing and termination detection — coupled with details on the messiness of the implementation and the compromises needed to implement user-friendly features like checkpoint and progress reporting.
At lunch, CW finally explained the green dot on my badge and on CM's badge: it doesn't indicate a site that hasn't paid; instead it's a sign that the owner is a vegetarian — not that the veggie lunches were anything to write home about.
The afternoon was application focused and I made a point of attending both LA's and MG's presentations, both of which were good although I'm not sure I'm ever going to understand plasma physics. I followed this with a talk on drain times which was interesting but which relied on output from Moab to implement.
The last couple of sessions were focused on SSDs, firstly an overview of Intel's new NVMe devices, followed by a presentation by DDN on their Infinite Memory Engine — essentially a neat way of placing a burst buffers in front of standard parallel file systems to improve the performance of oddly shaped IO requests.
And with that the conference was essentially complete.
The technical sessions were excellent, I talked to a lot of people — but I'm acutely aware there were lots I missed — and it was good to catch up with people I haven't seen for ages. I've got lots of useful ideas to take home with me and it's inspired me to do more fundamental stuff over the next year and to try to put together a paper for 2017.
