Trac July 2011

trac@lists.einsteintoolkit.org

2 participants
130 discussions

[Einstein Toolkit] #479: speed up VisIt's CarpetHDF5 plugin, set 2d masks
by Einstein Toolkit 25 Sep '11

25 Sep '11

#479: speed up VisIt's CarpetHDF5 plugin, set 2d masks -------------------+-------------------------------------------------------- Reporter: rhaas | Type: enhancement Status: new | Priority: minor Milestone: | Component: Other Version: | Keywords: CarpetHDF5 -------------------+-------------------------------------------------------- Hello all, the attached patches (when all of them are applied to VisIt's CarpetHDF5 plugin) speed up opening (large) HDF5 output files in VisIt by about a factor of 8. The two main speedups are replacing a linear search when translating from the Cactus iteration cctk_iteration to timestep (index) number and (surprisingly enough) the parsing of the Cactus variable name out of the dataset name rather than the "name" attribute (but it falls back to reading the attribute if the parsing fails). The third speed improvement is only visible when more than one file are opened and plotted. VisIt seems to consider reading metadata a cheap operation and creates and destroys the metadata object (avtCarpet...) very often. The patch caches metadata when VisIt destroys the object. This speeds up plotting several frames (or using the timestep slider) considerably, but has the unfortunate side effect that one cannot fully close files anymore (re-loading still works though). I attach timing information to show the gains. Data files and a python script to use with visit -cli -s openfile.py are provided at http://www.numrel.org/~rhaas3/CarpetHDF5/ . The data files are about 20MB when compressed and about 2-4GB when decompressed (they are actual data from one of my simulations with all values set to zero). There are three data sets provided. "flat" is the direct output from the Cactus simulation. "grouped" contains datasets from each timestep in a group of their own (grouping is mostly useful when mergin hdf5 files which is unbearably slow otherwise). "unchunked" used a python script to merge the individual Carpet components into the CarpetRegrid2 boxes (or equivalent), which results in much faster load times and smaller files (but is itself a slow operation). Code and scripts to group and unchunk hdf5 data is not yet public. If there is interested and I can post those as well but they are not nicely coded at all. The last patch (actually the first two by number) changes the way CarpetHDF5 sets up the nesting of components for 2D data so that eg. contour plots work properly (no more duplicate lines from the coarse points under fine points). -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/479> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 2

[Einstein Toolkit] #459: Autodetect sched_getaffinity
by Einstein Toolkit 25 Sep '11

25 Sep '11

#459: Autodetect sched_getaffinity -------------------------+-------------------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: Cactus | Version: Keywords: | -------------------------+-------------------------------------------------- The attached patch autodetects whether sched_getaffinity is available. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/459> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 4

[Einstein Toolkit] #443: Output a stack backtrace on a fatal error in a simulation
by Einstein Toolkit 21 Sep '11

21 Sep '11

#443: Output a stack backtrace on a fatal error in a simulation -------------------------+-------------------------------------------------- Reporter: hinder | Owner: eschnett Type: enhancement | Status: new Priority: major | Milestone: Component: Carpet | Version: Keywords: | -------------------------+-------------------------------------------------- When a Cactus simulation aborts with a signal, it is often difficult to determine which part of the code led to the problem. The attached patch registers a signal handler on Carpet startup for signals 11 and 6 (segmentation fault and abort, e.g. from assert()) which outputs a stack backtrace from each process to a file, including demangling symbol names. It uses some low-level and possibly unofficial APIs, and is likely not completely portable. However, I have tested it on Mac OS (gcc) and Linux (intel) and it works in those places. Part of this code was contributed by Justin Luitjens at the Carpet developers' workshop in summer 2010. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/443> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 16

[Einstein Toolkit] #259: Update PETSc to 3.1p7
by Einstein Toolkit 20 Sep '11

20 Sep '11

#259: Update PETSc to 3.1p7 -------------------------+-------------------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: Cactus | Version: Keywords: | -------------------------+-------------------------------------------------- The current version of 3.1p5, the difference are probably only small bug fixes. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/259> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 1

[Einstein Toolkit] #401: pysim fails to build a new configuration because of old data
by Einstein Toolkit 14 Sep '11

14 Sep '11

#401: pysim fails to build a new configuration because of old data ------------------------+--------------------------------------------------- Reporter: knarf | Owner: mthomas Type: defect | Status: new Priority: major | Milestone: ET_2011_11 Component: SimFactory | Version: Keywords: | ------------------------+--------------------------------------------------- I just tried 'sim build --thornlist=./thornlists/Cactus.th' and it failed: Info: Simfactory command: /nics/d/home/knarf/utils/simfactory/bin/../lib/sim.py "build" "-- thornlist=./thornlists/Cactus.th" Info: Version 1318 The Simulation Factory: Manage Cactus simulations Info: defs: /nics/d/home/knarf/utils/simfactory/etc/defs.ini Info: defs.local: /nics/d/home/knarf/utils/simfactory/etc/defs.local.ini Warning: properties.ini does not exist for simulation IOL1_CCT_whisky_tov_1_1 Traceback (most recent call last): File "/nics/d/home/knarf/utils/simfactory/bin/../lib/sim.py", line 142, in ? main() File "/nics/d/home/knarf/utils/simfactory/bin/../lib/sim.py", line 130, in main restartlib.CleanupRestarts() File "/nics/d/home/knarf/utils/simfactory/lib/restartlib.py", line 165, in CleanupRestarts sim_create_time = restart.GetMarkTime("simulation") File "/nics/d/home/knarf/utils/simfactory/lib/simrestart.py", line 65, in GetMarkTime for fptr in os.listdir(self.InternalDir): OSError: [Errno 2] No such file or directory: '/lustre/scratch/knarf/simulations/IOL1_CCT_whisky_tov_1_1/SIMFACTORY' It should not fail to build because of old data - regardless of how that looks like. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/401> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 3

[Einstein Toolkit] #440: Cannot check out Carpet non-anonymously
by Einstein Toolkit 14 Sep '11

14 Sep '11

#440: Cannot check out Carpet non-anonymously ---------------------------+------------------------------------------------ Reporter: eschnett | Owner: eric9 Type: defect | Status: new Priority: major | Milestone: Component: GetComponents | Version: Keywords: | ---------------------------+------------------------------------------------ Whenever I try to check out Carpet non-anonymously using the standard thorn list, I receive the error message $ ./bin/GetComponents --update --root=. manifest/einsteintoolkit.th Error: The URL for Carpet/doc has changed, please perform a clean checkout. (This is not a helpful error message, because the urls are not shown.) I find that, for Mercurial, GetComponents always sets $url = $component{"URL"}, using the anonymous url. Later it compares this url to the url actually used to check out the repository, which is in my case different, since I used the non-anonymous, authorized url. This leads to an unconditional error. The file $HOME/.crl/users does not enter, since I never enter a password for this repository. The only way to circumvent this problem is to modify the URL in the thorn list to make it identical to the AUTH_URL. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/440> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 5

[Einstein Toolkit] #406: publication list is shuffled
by Einstein Toolkit 14 Sep '11

14 Sep '11

#406: publication list is shuffled -------------------------------------+-------------------------------------- Reporter: bmundim | Owner: Type: defect | Status: new Priority: major | Milestone: ET_2011_11 Component: EinsteinToolkit website | Version: Keywords: | -------------------------------------+-------------------------------------- The publication list at http://einsteintoolkit.org/publications/ is said to be arranged by year, but at the moment it seems to be completely shuffled. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/406> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 3

[Einstein Toolkit] #256: Add support for the new Lonestar
by Einstein Toolkit 09 Sep '11

09 Sep '11

#256: Add support for the new Lonestar -------------------------+-------------------------------------------------- Reporter: eschnett | Owner: mthomas Type: enhancement | Status: new Priority: minor | Milestone: Component: SimFactory | Version: Keywords: | -------------------------+-------------------------------------------------- -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/256> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 2

[Einstein Toolkit] #108: LocalInterp and LocalReduce should move to CactusNumerical
by Einstein Toolkit 06 Sep '11

06 Sep '11

#108: LocalInterp and LocalReduce should move to CactusNumerical -----------------------+---------------------------------------------------- Reporter: anonymous | Type: enhancement Status: new | Priority: minor Milestone: | Component: Cactus Version: | Keywords: -----------------------+---------------------------------------------------- -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/108> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 3

[Einstein Toolkit] #203: Add shallow clone to hg
by Einstein Toolkit 03 Sep '11

03 Sep '11

#203: Add shallow clone to hg ---------------------------+------------------------------------------------ Reporter: eric9 | Owner: eric9 Type: defect | Status: new Priority: minor | Milestone: Component: GetComponents | Version: Keywords: | ---------------------------+------------------------------------------------ Git and darcs offer a shallow clone option (--depth 1 for git and --lazy for darcs). Hg can accomplish a similar thing by running hg clone -r tip URL. This should be implemented following the same rules as shallow clones for git and darcs. -- Ticket URL: <https://trac.einsteintoolkit.org/ticket/203> Einstein Toolkit <http://einsteintoolkit.org> The Einstein Toolkit

1 5

← Newer
1
...
6
7
8
9
10
11
12
13
Older →

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

Trac July 2011