I consider that this blog belongs in part to the HPC arena. I don't blog too often about it though. Doesn't that seem a bit of a contradiction? Well, kinda, sorta.
The one of the reasons that I don't do it too often is because of the issues wrt to all the NDAs I sign. I sign a helluvalota them. Stuff that slashdot and others hear about or that the press gets 'leaked' to I normally hear about two years ahead of time. By then, to me its old hat. A lot of people are going on about something someone has planned and is bringing to market when we were briefed some time ago and asked for our feedback.
The second reason is office politics. When you have a lot of double plus tech enabled geeks, they're bound to find what I write online. Sometimes they do. I know that at least two of my coworkers read this blog. I'd rather not say something that will get me gobsmacking nuked at work. That nearly happened when a series of posts were tangentially related to work.
Th third reason happens to be inter-site politics. There are a few sites that are not so much collaborators as outright rivals and enemies. Funding raids are not unheard of. HPC is a small stream with lots of cash flowing through it. It has to be: these machines are fscking expensive. Often large upgrades in processing power are achieved once you have crossed a certain dollar threshold: frex, if you pay $30 million for your Unfundia Mandatium UberFlopPuter-7 and then can get an additional $5 million, you get more than a 1/6 increase in processing powerr and sustained performance. Reason being that there are infrastructure bits that all machines need that are often paid for in that first chunk (disk space, interconnect, etc) that are easily scalable (to some extent) with a little more money. Lab X then wants to get that money and sometimes Lab Y has it but looks vulernable. I'd rather not be the reason that we would EVER look vulernable! And we're not. We're actually rather dangerous in our own right.
When I do post about HPC, it's normally about a side project I am working on that isn't so important to the shape of the supercomputing world. Often my side projects have some serious implications for those that are at the top and take time to think, but they're emphatetically not stuff that will impact our funding.
I do have a project that I am cooking. I need to find a stylized terror bird skull logo first. I will need to generate 120 TB of data - I insist on real science data instead of just goofy, nasty bits that will be thrown away later. That means a science app - and project - that generates that much data. I am going to need a terabit fat pipe. and the associated fiber. The disk space too: that's not too hard, I have a vendor that's itching to provide this. I need some servers and a few HPC platforms (check) (and yes, few). Most importantly, I need management sign-up.
That is going to be harder than all the rest.
We'll see what we can do. The question is whether or not to do it in SC07 or SC08 (Reno ro Austin respectively). Time will tell...and time will tell whether or not I will get to do this at all.
The one of the reasons that I don't do it too often is because of the issues wrt to all the NDAs I sign. I sign a helluvalota them. Stuff that slashdot and others hear about or that the press gets 'leaked' to I normally hear about two years ahead of time. By then, to me its old hat. A lot of people are going on about something someone has planned and is bringing to market when we were briefed some time ago and asked for our feedback.
The second reason is office politics. When you have a lot of double plus tech enabled geeks, they're bound to find what I write online. Sometimes they do. I know that at least two of my coworkers read this blog. I'd rather not say something that will get me gobsmacking nuked at work. That nearly happened when a series of posts were tangentially related to work.
Th third reason happens to be inter-site politics. There are a few sites that are not so much collaborators as outright rivals and enemies. Funding raids are not unheard of. HPC is a small stream with lots of cash flowing through it. It has to be: these machines are fscking expensive. Often large upgrades in processing power are achieved once you have crossed a certain dollar threshold: frex, if you pay $30 million for your Unfundia Mandatium UberFlopPuter-7 and then can get an additional $5 million, you get more than a 1/6 increase in processing powerr and sustained performance. Reason being that there are infrastructure bits that all machines need that are often paid for in that first chunk (disk space, interconnect, etc) that are easily scalable (to some extent) with a little more money. Lab X then wants to get that money and sometimes Lab Y has it but looks vulernable. I'd rather not be the reason that we would EVER look vulernable! And we're not. We're actually rather dangerous in our own right.
When I do post about HPC, it's normally about a side project I am working on that isn't so important to the shape of the supercomputing world. Often my side projects have some serious implications for those that are at the top and take time to think, but they're emphatetically not stuff that will impact our funding.
I do have a project that I am cooking. I need to find a stylized terror bird skull logo first. I will need to generate 120 TB of data - I insist on real science data instead of just goofy, nasty bits that will be thrown away later. That means a science app - and project - that generates that much data. I am going to need a terabit fat pipe. and the associated fiber. The disk space too: that's not too hard, I have a vendor that's itching to provide this. I need some servers and a few HPC platforms (check) (and yes, few). Most importantly, I need management sign-up.
That is going to be harder than all the rest.
We'll see what we can do. The question is whether or not to do it in SC07 or SC08 (Reno ro Austin respectively). Time will tell...and time will tell whether or not I will get to do this at all.
No comments:
Post a Comment