home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.graphics.avs
- Path: sparky!uunet!paladin.american.edu!howland.reston.ans.net!usc!sol.ctr.columbia.edu!aurora!larryg
- From: larryg@avs.com (Larry Gelberg)
- Subject: Big Dataset Survey Results
- Sender: nobody@ctr.columbia.edu
- Organization: Advanced Visual Systems Inc.
- Date: Mon, 25 Jan 1993 19:12:36 GMT
- X-Newsreader: Tin 1.1 PL4
- Message-ID: <1993Jan25.191236.19800@sol.ctr.columbia.edu>
- X-Posted-From: aurora.avs.com
- NNTP-Posting-Host: sol.ctr.columbia.edu
- Lines: 209
-
- Here are the results of the Big Dataset Survey I queried the network
- about several weeks ago. You may remember that I asked the question
- "What is the biggest dataset YOU have ever pushed through AVS?" The
- numbers are interesting for several reasons: the sizes of the datasets
- presented, and the variety of techniques for which AVS is being used.
-
- Here is a summary table followed by the brief descriptions everyone
- sent me. Please continue to send me this information - I will
- continue to compile it and present the results at the User's
- Group Meeting in May!
-
- Fields:
- 67,108,864 Bytes - Marc Kessler, U of Mich. Hospital
- - (Anonymous)
- 60,000,000 Bytes - Ronald J.G. Goossens, Stanford University
- 51,800,000 Bytes - Wade K. Smith, Convex Computer Corporation
- 24,576,000 Bytes - Jamie Pelagatti, Harvard Medical School
- 20,961,952 Bytes - Ron Karwoski, BIR, Mayo Foundation
- 12,582,912 Bytes - Larry Gelberg, Advanced Visual Systems Inc.
- 8,388,608 Bytes - Pierre MARTEL, C N E T PAA/TIM/SSI
- 3,000,000 Bytes - Rob Lake, BP Research
- 120,000 Bytes - Andrew Rhodes, MIT
-
- UCD:
- 262,144 cells - Alexander Yarmarkovich, Advanced Visual Systems Inc.
- 100,000 cells - Peter Yerburgh, Intera
- 16,405 cells - John Sheehan, Advanced Visual Systems Inc.
-
- ----------------------------------------------------------------------
-
- 67,108,864 Bytes
- 256x256x128 doubles - 3D convolutions
- Stellar GS with 128M
- Marc Kessler, University of Michigan Hospital
-
- ----------------------------------------------------------------------
-
- 67,108,864 Bytes
-
- I've done volume rendering of a 512x256x128 uniform integer volume
- on a XXXXX with XXXXX, and my volume renderer runs
- in about XXXXX seconds. At this point the results are too
- preliminary to be publishable (since I just wrote the renderer).
- But I'll let you know when the results are publishable
- (probably about the time we release the module).
-
- XXXXX, XXXXX (wishes to remain anonymous until results are
- officially published)
- ----------------------------------------------------------------------
- 60,000,000 Bytes
-
- We have a 3D semiconductor simulation code that models the electrical
- behavior of submicron transistors. It solves the Poisson equation and
- two transport equations. The output consists of electrical potential,
- electron and hole concentration, electron and hole current density at
- each node in the computational mesh, i.e. 9 double-precision
- floats/grid point. The code is called STRIDE and uses rectangular
- blocks as mesh elements.
-
- We are involved in a number of high-performance computing efforts and
- have been running STRIDE on an in-house Intel hypercube. Early 1992 we
- got access to the Intel DELTA machine at Caltech in Pasadena, CA.
- There, we were able to analyse a full series of electrical biases (13
- different solutions) on a grid of 5 million points (approximately
- 150x150x210). Any scalar solution variable represented 20Mb per bias
- point, vector variables are 60Mb per bias point. The total solution
- file contained 13x9x4x5M bytes = 2.34 Gbyte.
-
- These data were transferred to the National Center for Computational
- Electronics at the University of Illinois. The most interesting visual
- corresponded to a 3D streamline rendition of hole current for one of
- the bias points, with a color contour map of the magnitude of hole
- current for the central plane. That data set in itself contained 60
- Mbyte of data.
-
- I think - but am not sure - that the visualization was done on a
- Convex machine with 2Gbyte of main memory. It is one of Convex's
- high-end versions. The streamline visualization took about 16 hours of
- CPU time. The visualization was done "distributed" in the sense that
- we had hooked up to display windows to the geometry viewer, one
- locally at Illinois, the other at Stanford.
-
- The simulation and visualization received an honorable mention in the
- Intel Grand Challenge Award, which was organized last year to
- commemorate the first anniversary of the DELTA machine.
-
- Ronald J.G. Goossens
- Integrated Circuits Lab, Stanford University
-
- ----------------------------------------------------------------------
-
- 51,800,000 Bytes
- UCI cell data set (courtesy of D.B. Menzel, et. al.)
- 360x360x100 float field (51.8Mb) converted to byte (12.9Mb)
- Convex C220 (2 heads)
- 512Mb physical memory
- gradient shade 144 seconds
- tracer 14 seconds
- isosurface plus render (grew to 800Mb) 30 minutes
- work done by Duane Gustavus at Convex
-
- Wade K. Smith, Convex Computer Corporation
-
- ----------------------------------------------------------------------
-
- 24,576,000 Bytes
- We have sent 320x320x60 integer (i.e 4 byte) data sets through
- AVS networks. These are CT data sets typically. We've used a
- Stardent GS2000 with 128 Mb of memory. Usually, we read it
- then slice it or byte-reduce it before rendering it. Just
- reading it into a field did not take more than a few minutes.
- It will also flow through the pipes with a couple of minutes too.
-
- Jamie Pelagatti, Harvard Medical School
-
- ----------------------------------------------------------------------
-
- 20,961,952 Bytes
- I have loaded a 20 Mb file, 226 x 272 x 341, (byte field) on a
- HP 9000/720 with 64 Mb of memory. The data is a CT scan of the
- head which we interpolated to make cubic. After the initial
- reading in of the data I am able to render and display the
- data in ~5 seconds. Render types include: Voxel Gradient
- Shading, Maximum Intensity Projections, and Summed
- Projections. Threshold values and three rotation angles (x, y,
- z) can be set by the user.
-
- Read ANALYZE->ANALYZE Render->Colorizer->Display Image
-
- Read ANALYZE and ANALYZE Render are our own AVS Modules which
- have limited ANALYZE functionality.
-
- I'm sure we can go larger, I don't have any larger datasets handy
- to find the limit.
-
-
- Ron Karwoski Internet: rak@bru.mayo.edu
- Biomedical Imaging Resource, Mayo Foundation
- ----------------------------------------------------------------------
-
- 12,582,912 Bytes
- 256 x 256 x 192 byte volume
- Volume Rendering using TRACER and COMP SHADE
- Stellar GS 2000 w/ 128 MBytes
-
- Larry Gelberg, Advanced Visual Systems
- ----------------------------------------------------------------------
-
- 8,388,608 Bytes
- It takes near 3 minutes for a 3D float field with color 524288
- particules with a "scatter dot" module.
-
- 524,288 Bytes
- It takes near 2 secondes for a 2D float field 128x1024 throught
- the "field to mesh" module.
-
- 500,000 Bytes
- It takes near 5 secondes for a 3D float field 50x50x50 with an
- "isosurface" module.
-
- and, of course, all these modules are connected out a "render geometry".
- I'm using AVS on a Stardent 800 with 32 Mbytes of memory.
-
- Pierre MARTEL, C N E T PAA/TIM/SSI
-
- ----------------------------------------------------------------------
-
- 3,000,000 Bytes (1000 * 250 * 4 (sizeof float) * 3 ([XYZ]))
- Field 2-D, some 3-D Fields up to 1000 by 250 or so. Now using
- 3-D topo map about 175 by 150, rendered with 608 by 865 image.
- KPC Titan 3000VS dualprocessor - 128MB
- At 60,000 to 100,000 triangles in the generated geometry, rotation
- and scaling are at the limit of usability. 3-D topo map quite
- usable interactively.
-
- Rob Lake, BP Research
-
- ----------------------------------------------------------------------
-
- 120,000 bytes (10000 * 4 (sizeof float) * 3 ([XYZ])
- We were displaying surfaces (as geometries) comprised of 10,000 points
- at a pretty good speed. For reasons I won't get into, we actually created
- text files N x 10000 lines long with N surfaces in them. The module that
- reads the files used a direct positioning which worked nearly instantly.
- We did find that jumping up to 15,000 points crossed a threshold and led
- to big slowdowns -- likely we ran out of memory and it had to swap.
-
- Our system is a DECstation 5000/200 with Ultrix 4.2, DECwindows, and AVS 4.0.
- The system has 40 Meg RAM, 160 Meg swap space.
-
- Andrew Rhodes, MIT
-
- ----------------------------------------------------------------------
-
- UCD:
-
- 262,144 cells Alexander Yarmarkovich, Advanced Visual Systems Inc.
- (hydrogen.dat converted to cells for testing purposes)
- 100,000 cells Peter Yerburgh, Intera
- 16,405 cells John Sheehan, Advanced Visual Systems Inc.
-
- ----------------------------------------------------------------------
-
-
- --
- === Larry Gelberg ============================ larryg@avs.com =======
- Advanced Visual Systems Inc. (AVS Inc.)
- 300 Fifth Ave, Waltham, MA 02154
- ===== Tel: 617-890-4300 = Fax: 617-890-8287 =========================
-