siCluster decloaking

The day job pushed the announcement which went into the SC09 PR black hole out the door today. Also, the first installation is also indicated in the next announcement … Scalable Informatics Introduces siCluster, an Innovative and Highly Scalable Performance Storage Cluster Canton, MI, Dec 2, 2009 – Scalable Informatics Inc., a provider of innovative … Read moresiCluster decloaking

I am not sure I should be amused, but I am …

The day job has a registration page for people who are interested in purchasing high performance storage, deskside supercomputing, storage clusters, and other HPC like things from us.
This registration page asks some very simple things: who are you, where are you, what your shipping/billing address is, what you want your user name to be, and what things you want information on.
Registration is, fundamentally, a matter of trust. We guarantee we will not spam people. The entire purpose of this is NOT to gather marketing information. It is to let people order from our site, and get products from us online.
I respect people who are leery of filling this stuff out. They think (based upon past experience with other groups) that they will be spammed.
I should be quite unequivocal about this. No. This isn’t marketing information gathering for email lists. We treat the users EXACTLY the way we, and me personally (since I run the company, its easy for me to set this policy), would like to be treated. With respect. Without unnecessary emails/contacts.
Further, we handle obviously false data in accounts by deleting the account. We don’t have time to waste with garbage like this, so we handle it in the only reasonable manner. Once you have an account, you can order online. This helps tremendously in fraud prevention.
Ok so far … now here is what I know I shouldn’t find amusing, but I do.

Read moreI am not sure I should be amused, but I am …

Storage cluster drag racing …

… well, I am trying to figure out what I am doing wrong in io-bm. I need a new method to defeat some of the smarter caching bits, my MPI_Send/MPI_Recv pairs are blocking pairs, and this impacted performance. Not only that, the additional traffic over the Infiniband was definitely a cause of contention on the wire.
Doing some TB sized writes at a good rate. The “naive” bandwidths (the way IOzone calculates them) are about where we predicted given the measured IB performance.
Will report more soon, expect a white paper and other docs soon.
As a teaser, here is a 128 GB read. The longest of the 128 threads took 12.11 seconds.

Read moreStorage cluster drag racing …