My labmates and I (Rick Wagner in particular) spent the last week feverishly building the system. The primary work was to build the monitor rack. It has many bolts and sliders and adjustments. We arranged to have two holes drilled through the wall for the monitor cables. The 30-inch screens require a very particular kind of cable and getting enough of the right kind that were long enough to reach all the screens was surprisingly difficult.

My main task was to convert some of the publicly available high-resolution astronomical images into tiled TIFFs. Tiled TIFFs break the data up into easy to scale tiles, which is essential for the optiportal. I used many of the images from this page, including the 403 megapixel Carina Nebula image linked there. It was quite frustrating. I tried half a dozen computers and iterations of ImageMagick and VIPS before I finally got a combination that worked.

In some ways, this is worse than a projector. There are gaps between the screens. This requires multiple expensive computers and many, many cables. However, there are some big advantages to this setup. There is no projector that has anywhere the same resolution as this optiportal (2560×1600x20 = 81.92 megapixels). Just one of the monitors is better than all but the most specialized and expensive projectors. The screens are bright and clear, even in normal lighting conditions. Unlike a projector, standing in front of the screen doesn’t shadow the image. It has to be seen for yourself; twenty 30-inch monitors are awesome to behold.

Each four-monitor column is powered by a HP xw8600 workstation with one 4-core 2.33GHz Xeon processor, 4GB of RAM, two NVIDIA Quadro FX4600 768MB video cards with dual output, running Ubuntu 8.04 LTS. There is a sixth identical workstation that serves as a head node which sits in front of the screens and controls the output on the wall. There is also a HP ML370 server that has a 1TB RAID5 disk array, and has room for an additional 1TB. Eventually, all the machines will be connected by a 10Gb CX-4 network using a ProCurve 6400cl-6XG switch, and the head node and disk server will be connected to the outside world with a 10Gb optical connection. The monitor array is controlled using CGLX which is developed at Calit2, which is here at UCSD.

Science will not be performed on the optiportal, at least the major calculations won’t. Instead, visualizations of simulations run on supercomputers will be displayed and analyzed. The human eye and brain is still a superior judge of the accuracy and meaning of data than a set of conditional equations. It is often necessary to actually see the raw data as clearly as possible. The point of the eventual 10Gb optical connection is that various parallel visualization programs like VisIt and ParaView will be practical. These programs do the data analysis on a remote supercomputer and the visualization is sent to the optiportal in real time for display. In this way, terabytes of raw data don’t have to be transferred off the supercomputer, and the computationally intensive part of the visualization takes place on the supercomputer, which has far more processors, RAM and disk space than the optiportal. I hope to someday have some of my data displayed in this fashion.

I’ve put up some pictures and movies in my gallery, and I encourage you to check it out. There’s a really fun roller coaster animation I think you’ll like.

What you see below is one job running 8192 cores (a), another running on 4096 (h), one with 2048 (k) and a smattering of smaller jobs. My jobs are i and j, running on 8 cores each. The computer is just about full here, about 96% usage.

This also allows me to know who to blame when my jobs are sitting waiting to start for days.

Compute Processor Allocation Status as of Fri Aug  1 18:14:16 2008
 
     C0-0     C0-1     C0-2     C0-3     C1-0     C1-1     C1-2     C1-3     
  n3 -------- -------- hhhhhhhh hhhhhhhh SSSaaaaa aaaaaaaa aaaaaaaa aaaaaaak 
  n2 -------- -------- hhhhhhhh hhhhhhhh    aaaaa aaaaaaaa aaaaaaaa aaaaaaak 
  n1 -------- -------- hhhhhhhh hhhhhhhh    aaaaa aaaaaaaa aaaaaaaa aaaaaaak 
c2n0 -------- -------- hhhhhhhh hhhhhhhh SSSaaaaa aaaaaaaa aaaaaaaa aaaaaaak 
  n3 SSSSS--- -------- hhhhhhhh hhhhhhhh SSSSaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2      --- -------- hhhhhhhh hhhhhhhh     aaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1      --- -------- hhhhhhhh hhhhhhhh     aaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c1n0 SSSSS--- -------- hhhhhhhh hhhhhhhh SSSSaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 SSSSSSSS -------- --lihhhh hhhhhhhh SSSSaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2          -------- --ljhhhh hhhhhhhh     aaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1          -------- --ljhhhh hhhhhhhh     aaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c0n0 SSSSSSSS -------- ---lihhh hhhhhhhh SSSSaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
    s01234567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 
 
     C2-0     C2-1     C2-2     C2-3     C3-0     C3-1     C3-2     C3-3     
  n3 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c2n0 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c1n0 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c0n0 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
    s01234567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 
 
     C4-0     C4-1     C4-2     C4-3     C5-0     C5-1     C5-2     C5-3     
  n3 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c2n0 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c1n0 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c0n0 hhhhhhhh hhhhhhhh hhhhhhhh hhhhhhhh aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
    s01234567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 
 
     C6-0     C6-1     C6-2     C6-3     C7-0     C7-1     C7-2     C7-3     
  n3 hhhhcccc cccccccc bbbbbbbb gggggggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhcccc cccccccc bbbbbbbb gggggggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhcccc cccccccc bbbbbbbb gggggggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c2n0 hhhhhccc cccccccc bbbbbbbb gggggggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 hhhhhhhh cccccccc bbbbbbbb bbbbgggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh cccccccc bbbbbbbb bbbbgggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh cccccccc bbbbbbbb bbbbgggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c1n0 hhhhhhhh cccccccc bbbbbbbb bbbbbggg aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 hhhhhhhh cccccccc ccccbbbb bbbbbbbb aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 hhhhhhhh cccccccc ccccbbbb bbbbbbbb aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 hhhhhhhh cccccccc ccccbbbb bbbbbbbb aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c0n0 hhhhhhhh cccccccc cccccbbb bbbbbbbb aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
    s01234567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 
 
     C8-0     C8-1     C8-2     C8-3     C9-0     C9-1     C9-2     C9-3     
  n3 ggggffff ffffffff eeeeeeee dddddddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 ggggffff ffffffff eeeeeeee dddddddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 ggggffff ffffffff eeeeeeee dddddddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c2n0 gggggfff ffffffff eeeeeeee dddddddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 gggggggg ffffffff eeeeeeee eeeedddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 gggggggg ffffffff eeeeeeee eeeedddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 gggggggg ffffffff eeeeeeee eeeedddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c1n0 gggggggg ffffffff eeeeeeee eeeeeddd aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n3 gggggggg ffffffff ffffeeee eeeeeeee aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n2 gggggggg ffffffff ffffeeee eeeeeeee aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
  n1 gggggggg ffffffff ffffeeee eeeeeeee aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
c0n0 gggggggg ffffffff fffffeee eeeeeeee aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa 
    s01234567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 
 
     C10-0    C10-1    C10-2    C10-3    C11-0    C11-1    C11-2    C11-3    
  n3 ddddkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
  n2 ddddkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
  n1 ddddkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
c2n0 dddddkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
  n3 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
  n2 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
  n1 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
c1n0 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk aaaaaaaa aaaaaaaa 
  n3 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkXkkk kkkkkkkk kkkkaaaa aaaaaaaa 
  n2 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkaaaa aaaaaaaa 
  n1 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk Xkkkkkkk kkkkaaaa aaaaaaaa 
c0n0 dddddddd kkkkkkkk kkkkkkkk kkkkkkkk kkkkkkkk kkkkXkkk kkkkaaaa aaaaaaaa 
    s01234567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 
 
Legend:
   nonexistent node                 S  service node
;  free interactive compute CNL     -  free batch compute node CNL
A  allocated, but idle compute node ?  suspect compute node
X  down compute node                Y  down or admindown service node
Z  admindown compute node           R  node is routing
 
Available compute nodes:       0 interactive,   149 batch
 
ALPS JOBS LAUNCHED ON COMPUTE NODES
Job ID     User       Size   Age              command line
--- ------ --------   -----  ---------------  ----------------------------------
 a  155793 xxxxxx      2048  9h00m            xxxxxxxx
 b  156058 xxxxxxxx     128  0h50m            xxxxxxxx
 c  156060 xxxxxxxx     128  0h50m            xxxxxxxx
 d  156062 xxxxxxxx     128  0h49m            xxxxxxxx
 e  156064 xxxxxxxx     128  0h49m            xxxxxxxx
 f  156066 xxxxxxxx     128  0h48m            xxxxxxxx
 g  156068 xxxxxxxx     128  0h48m            xxxxxxxx
 h  156080 xxxxxxx     1024  0h33m            xxxxxxxx
 i  156085 sskory         2  0h22m            enzo.exe
 j  156087 sskory         2  0h21m            enzo.exe
 k  156089 xxxxxxxx     512  0h04m            xxxxxxxx
 l  156091 xxxx           4  0h02m            xxxxxxxx

(* I’m pretty sure that’s the layout. I could be wrong, so don’t invade a medium-sized oil-producing country based on that intelligence.)

A few months ago, Ranger was turned on. It is a Sun cluster in Texas with 63,000 Intel CPU cores. It is currently ranked fourth fastest in the world. Datastar has only 2528 CPUs (but those are real CPUs, while Ranger has mutli-core chips which in reality aren’t as good). By raw numbers, Ranger is an order of magnitude better than Datastar, except that Ranger doesn’t work very well. Many different people are seeing memory leaks using vastly different codes. These codes work well on other machines. I have yet to be able to run anything at all on Ranger. For all intents and purposes, Ranger is useless to me right now.

If you look at the top of the list of super computers, you’ll see that a machine called Roadrunner is the fastest in the world. Notice that it is made up of both AMD Opteron and IBM Cell processors. The Cell processor is the one inside Playstation 3s. Having two kind of chips adds a layer of complexity, which makes the machine less useful. The Cell processor is a vector processor, which is only awesome for very specially written code. The machine is fast, except it’s also highly unusable. I don’t have access to it because it’s a DOE machine, but a colleague has tried it and says he got under 0.1% peak theoretical speed out of it. Other people were seeing similar numbers. No one ever gets 100% from any machine, but 0.1% is terrible.

A Kraken

Computers two and three on the list are DOE machines, so I don’t have access to them. On the near horizon is a machine called Kraken, in Tennessee. It’s being upgraded right now, but when it’s complete it will be very similar to, but faster than the fifth fastest computer on the list currently, called Jaguar. It is a Cray XT4 that runs AMD Opteron chips. I got to use Kraken recently while it was still an XT3, and it was awesome. Unlike Ranger, it actually works. As an XT4, it should be even faster than Ranger. It will also have a great tape backup system, unlike Ranger.

I am predicting that Kraken will be come my new favorite super computer, replacing Datastar. However, I think it’s a shame that Datastar is being turned off even though it’s still very useful and popular. When it’s turned off to make way for machines like Ranger and Roadrunner(*), that’s just stupid.

(*) The pots of money for Ranger, Datstar and Roadrunner are different, but you get the point. Supercomputers aren’t getting better; in some cases, they’re getting worse!

The graph above shows the speedup that a few OpenMP statements can give with very little effort. OpenMP is a simple way to parallelize a C/C++ program which allows you to run a program on many processors at once. However, unlike MPI which can run on many different machines (like a cluster), OpenMP can only be run on one computer at a time. Since most new machines have multiple processors (or cores), OpenMP is quite useful.

I’ve added a couple dozen OpenMP statements to the code I’m working on. The blue line shows how long (in seconds) it took me to run a test problem on between one and 32 processors. The green line shows the speedup compared to running on a single processor as a ratio of time. It is very typical of parallel programs that the speedup isn’t linear and flattens out at high thread count. This small test problem deviates at 16 processors; when I do a real run (which will be much larger and the parallelization more efficient) I may see nearly linear speedups all the way to 32 processors.

I think it’s pretty neat how with very little effort I was able to significantly speedup my code. If you have a little programming experience, you can take a look at some simple OpenMP examples and see for yourself just how easy OpenMP is.

— just notice that the two peaks are pretty much in the same places on both graphs, 1.5 and 2.2. The first graph shows physical data (stars) and a double-Gaussian fit (light solid line). The second graph is the result of my using Monte-Carlo fitting to make entirely fake data using the first curve. The real graph has over 10,000 items to make that smooth distribution, while with only about 100 items Monte-Carlo is already starting to look like the real thing. Of course, it will take much more items to capture the smoothness and the ‘long-tail’ on each end.

I just wanted to share because the whole thing I wrote, which includes a simple function integration (for normalization), worked on my first try.

Datastar is the 2500 CPU supercomputer I do most of my work on. Until today I had never seen it in person. Behind me are just a few of the racks that make up the computer. The room was loud and alternatively very hot and very cold. There are a few more pictures from the SDSC supercomputer room in my Cell Pics gallery.

In the first part of the movie, the cube is rotated about its center. Next, while looking along the z-axis, the volume of the cube plotted is reduced until only 1/10th of the cube is shown. Then, this 1/10th thickness is scanned through the entire cube, and then the volume plotted is replenished back to full.

The things to notice are how the dark matter form areas of high density, which are connected by filaments. Between these are areas of relatively few particles, which are called voids. This is how our universe really looks, with huge collections of galaxies clustered together, separated by huge expanses of nearly empty space. I should point out that there are many more galaxy haloes in this box besides the yellow boxes.

What I was looking for was one of those yellow boxes which is fairly isolated from areas of high density dark matter. I picked one, and now I am doing this same simulation again, but with higher resolution boxes centered on the area of interest. The simulations I’m doing right now are fairly cheap (in computer time currency). I want to be sure that when I run big, time-intensive simulations in the future that I’ve picked a good area to focus my attention on.

I did this visualization with Visit, a stereo, 4D visualization tool out of the Livermore National Lab. The ’stereo’ means that it can create two images of the same data that are slightly offset, which create a 3D effect if viewed correctly. The fourth dimension is for time, as it can handle time-ordered data sets. I then used Visit’s Python scripting features to output 800 individual PNGs, which I then stitched together (exactly like my time lapse movies) to make this movie.

Above is a part of what I’ve been working on lately. It’s a small part of the galaxy family tree that I’ve derived from a large simulation. In the simulation there are several ingredients that are thrown in the ‘box.’ Relevant to this are the dark matter particles, which coalesce into consituents of galaxies. The dark matter particles have unique id numbers. Using a some code I didn’t write, I process the simulation data and make a list of galaxies along with the particles in each galaxy. Then, using some code I did write, I track particles and galaxies over a number of time steps, which builds a relational mapping. Then, I use Graphviz to make a nice tree, as you see above.

Inside each box are either three or four data values. The top grid shows what percentage of the particles in that group came from no group, the middle grid shows both the number of dark matter particles that are in the group and the position of the group, while the bottom grid shows the percentage of the particles that go to no group. The simulation takes place inside of a 3D box with length 1 on a side and periodic boundaries (which means the distance between 0.9 and 0.1 is 0.2, not 0.8). The colors of the box correspond to its ranking in size, red is the largest, green the smallest. The numbers next to the arrows are what percentage of the parent group goes to the child group.

The goal of this is to get an idea of how the galaxies form over the course of the simulation. Of course the simulation tries to mirror reality, so this family tree may be worth something.