I’ve been chasing down a problem for a few days on a SLES 11 load. I’ve tried basic mdadm as well as the “Intel RAID”. Modified some of the mkinitrd scripts so that it doesn’t error out, and actually builds the initrd. But it never includes the mdadm or the /etc/mdadm.conf files. So the boot with the new initrd can’t assemble the raid correctly, and can’t do a correct switchroot to the raid device.

… sadly this appears to be true. Specifications were given, and we met the requirements, which entailed a demonstration of a particular level of performance over NFS. In case you aren’t sure, we demonstrated a sustained 1GB/s over NFS between 2 boxes over 10GbE last year. There aren’t too many companies that can do this. Our results were with RAID6 storage target, and an NFS client with small RAM size. Total read and write size each was much larger than either system memory.

I should define what I mean by marketing numbers. These are best effort benchmarking numbers assuming the best of all possible test cases, with equipment functioning solely for the benchmark test purposes. These are not benchmark results you will normally achieve in practice. They represent an extrema in performance. Measured benchmark numbers are sensitive to many factors. You need to perform several tests, make sure you can construct an “average” and make an assumption about the shape of the distribution around that average.

I’d mentioned this before in other contexts. But we just had an opportunity to bid on something with a very high data rate requirement. We provided a bid with a measurement indicating our performance, knowing full well we were one of very few vendors capable of this sort of performance. Yet purchasing folks appear not to take the “required” specifications into consideration. I actually feel bad for the user who is going to get a box that won’t be able to meet their needs, thanks to this process.

This was moving to Wordpress 2.9.2. One button. Thats it. Click it. Everything worked afterward. No drama, no complexity. No wizard. It. Just. Worked. I find this very inspirational. Makes me think we want to do this with our stack.

Raw end user accessible performance on data motion, data storage is rapidly becoming one of the most important problems in any HPC system. We’ve been talking about it for years, but its getting far more important by the day. And not just in HPC. I just spent a long time on the phone with someone from a government agency talking about their need for high performance storage, and analytical capability. We hear these refrains quite commonly, FC4/FC8 is simply too slow for their workloads, and they need to go faster.

We have been sent an RFP from a university we have some history with on bids. Our history has been, not winning the business. The winning bids sometimes (often) deviate wildly from the specifications as we read them. One thing I have learned from my experience with them is that the singular most important aspect of any bid is the price we present to them. You might think “well … duh” buts its more subtle than that.

I read an article on InsideHPC.com that I can’t say I agree with. The discussion on creative destruction is correct. You create a new market by destroying an old market. That has happened many times in HPC, by enabling better, cheaper, faster execution. If our SGI boxen of days old were 1/100th the cost of the Cray YMP at the time, and 20% of the performance, who won that battle? In all but a vanishingly small number of cases, SGI won.

Yeah … that was fun. Back at work. Things have been busy.

Spent several days in Orlando Florida (the house of mouse), and today drove to Key West, after a short stop in Miami to visit a friend, his wife, and their newly 1 year old daughter. Of course, upon getting to Key West, first our hotel door wouldn’t open, and then when we figured out the trick with the help of the hotel person, we inserted our card into the door, and the lights went out on the Island.