ECCE Version 1.5 Release Notes - December 1, 1999
Version 1.5 Patch #1 - February 1, 2000
Version 1.5 Patch #2 - May 17, 2000
Version 1.5 Patch #3 - September XX, 2000
Notes: It is critical that you read the sixth "What's new?" release note since starting ECCE has changed with the addition of a login window for security reasons.
It is not possible to use previous versions of ECCE after the release of version 1.5 due to changes in the database format.
The intent of this page is to provide information specific to version 1.5 of ECCE. Except as mentioned herein release notes from previous versions of ECCE still apply so please do not consider this as standalone documentation.
Release Notes for Recent Previous Versions
- Version V1.4.2 Release Notes - June 11, 1999
- Version V1.4.1 Release Notes - January 15, 1999
- Version V1.4 Release Notes - October 8, 1998
- (Sept. XX Patch)Globus 1.1.3 integration
- (Sept. XX Patch)Blue and pink molecular orbital colors
- (May 17 Patch)Molecular orbital colors
- (Feb. 1 Patch)Job monitoring reconnect capability
- (Feb. 1 Patch)Measure capability enhanced in Calculation Viewer
- (Feb. 1 Patch)Project database created on different platform warning
- (Feb. 1 Patch)NWChem ecce.out file appended to ecce.out.old for importing restarted calculations
- New login window and password encryption
- Symmetry Toolkit
- New measure capability in Calculation Viewer
- User preferences for atomic radii
- EMSL Collaboratory Electronic Lab Notebook integration
- Molecular orbital symmetry labels
- Enhancements to the Periodic Table
- Clipboard in Calculation Viewer
- Image sequence capture in Calculation Viewer
- Calculation Viewer property data "Options" menus
- NWChem MP2 "tight convergence" option supported
- nwtest and ecs1 registered as compute servers
- Project database format migration dialog
- Rearranged Builder main window layout
- Password encryption algorithm decoupled from ECCE distribution
- Project databases do not work between the Sun and SGI versions of ECCE V1.5
- Calculation Viewer property data panels get clipped
- (Sept. XX Patch)Machine configuration crash on SGI
- (Sept. XX Patch)Collaboratory Electronic Lab Notebook integration
- (Sept. XX Patch)Parsing large properties could fail
- (Sept. XX Patch)Remote shells did not come up due to incorrect DISPLAY environment variable
- (Sept. XX Patch)Job Launcher application crash when selecting machines
- (May 17 Patch)Remote xterm shells for mpp1 and ecs1
- (May 17 Patch)SGI passphrase entry deadlock
- (Feb. 1 Patch)Automatic scaling of the Calculation Viewer Normal Modes displacement vector magnitude
- (Feb. 1 Patch)Manage Files window "View" feature now uses $EDITOR
- (Feb. 1 Patch)Gaussian 98 on mpp1 and nwtest works with nproclinda
- (Feb. 1 Patch)Job Launcher Maximum Memory value correctly defaulted
- (Feb. 1 Patch)Babel PDB import exit code recognition fixed
- (Feb. 1 Patch)Potential for hanging on startup addressed
- Bond distance of created bonds
- Bond generation now works correctly
- CPK display
- Consistent "Display", "View", and "Options" menu items
- Bonding nubs now move with the surface of the atom
- Geometries for Gaussian 9X
- DFT calculations using NWChem
- Launches to NFS, AFS, and DFS calculation directories
- Better out of disk space detection when importing calculations
- Better end-of-file detection when importing calculations
- Better detection and handling of file transfer errors
- Periodic Table color and atomic radii preferences saved automatically
- Measures move after force field clean
- Babel import error detection improved
- 10 digit precision in input file
- Mulliken charge spectrum range
- Terminate using SIGTERM
- NWChem imports with coordinates in bohr
- i basis functions and NWChem
- ECCE_NWCHEM environment variable and parallel
- Project databases for old versions of ECCE can now be deleted
What's new? Top
- (Sept. XX Patch)The latest release of the Globus Toolkit, version 1.1.3, has been integrated into ECCE for job launches. Please see the Version V1.4.1 Release Notes for more information on Globus and the integration into ECCE. This latest version of Globus uses a patched version of ssh in order to support the Globus public key authentication model and allow shells for more than a single command. The lack of this multiple command shell was the reason ECCE has not supported Globus since V1.4.1 of ECCE. It is the responsibility of anyone wishing to use Globus for job launches to install both the Globus Toolkit clients and patched version of ssh on the local machine running ECCE. The Globus web site has the install directions and software download for the ssh patch. Additionally, the Globus server side daemons and a patched version of sshd must be available on the compute server.
- (Sept. XX Patch)The Calculation Viewer menubar Surface menu now has a blue and pink color combination option for molecular orbitals. This combination has been shown to work well for black and white publications when orbitals are rendered as a mesh instead of as a solid surface.
- (May 17 Patch)The Calculation Viewer menubar Surface menu allows the color combination for molecular orbitals to be changed. The default combination is still red and green. Red and blue along with blue and yellow combinations are now supported. For producing black and white images for publication there is also an option labeled "Light/Dark" which performs a grayscale rendering that better distinguishes the positive and negative lobes. The color combination selected also applies to "Positive/Negative" type cross-sectional slices although a slice must be hidden and then displayed again to show the new colors.
(Feb. 1 Patch)WARNING: The job monitoring reconnect capability described below
was designed for users running ECCE applications from a single workstation. If
you typically use more than one machine then please use the reconnect capability
with caution. This is unrelated to running the actual computational jobs on a
number of remote compute servers which is not an issue. If you do run ECCE
applications from different machines then it is your responsibility to know
whether the job monitoring is still active. It would be quite easy to
"reconnect" to a job that was already being monitored just from another machine.
In this case there would be duplicate data stored in the database as the job
would continue to be monitored on multiple machines. If you are certain that
you launched a particular calculation from the machine you are currently using
or that job monitoring has died on whatever machine it was started on
("ps -ef | grep eccejobstore" on the client machine helps you determine that)
then it is safe to reconnect.
Two new menu items in the Calculation Manager allow the checking and restarting of "job monitoring" done by ECCE. Every calculation (or job) run under ECCE has a pair of processes used to monitor output of the calculation, parse it, and populate the database. One process runs on the local workstation where the ECCE applications are running and one runs on the remote compute server where the calculation is being run. These processes run during the entire execution of the calculation even while idle in a queue. The process on the remote server communicates data properties from the output file to the process on the local machine which parses them and populates the database. There are a number of situations which can cause one or the other of these processes to fail. The most common are:
- Intended or unintended database outages shutting down the database connection
- Patches to the ECCE release causing executables and libraries to be replaced during job monitoring
- Various communication glitches that can happen over the duration of long running calculations causing the remote to local communication channel to go down
When failure does occur there will be no further updates to properties as displayed in the Calculation Viewer and the calculation will remain in the same state (either submitted or running) forever. To determine that job monitoring had failed without being familiar with the details of the job monitoring design in ECCE it was a matter of knowing that the calculation had actually finished based on the output file without the state of the calculation in ECCE updating.
The first new option in the Calculation Manager allows you to determine which submitted and running calculations aren't being monitored. Select the "Check All Job Monitoring" menu item from the Run Mgmt menu to determine if there are any calculations not being monitored in the entire project database. There will be a status message printed in the read-only text field at the bottom of the window. If there are calculations where job monitoring has failed then select the job and then the "Reconnect Job Monitoring" menu item in the Run Mgmt menu. This first clears the calculation of any properties already parsed and then restarts monitoring from the beginning of the output file on the remote server. If there is a significant amount of output already then it can take several minutes for the monitoring to "catch up" to the end of the output file where the calculation is currently at. If the Calculation Viewer is displaying a calculation where job monitoring is being reconnected then it will be extremely slow to respond to user input while it is rapidly updating properties almost appearing to be frozen. A calculation in the submitted state or with little output will very quickly catch up to the end of the current output. Within the Calculation Manager the state of the reconnected calculation will go back to submitted initially and quickly move to running if there are properties in the output file. If the calculation is already finished on the remote server then it will remain in the running state while the job monitoring is parsing the output file until it hits the end. The calculation will then change to the completed or incomplete state depending upon the exit status of the calculation.
The reconnect job monitoring feature can also be used to "re-parse" output from a previously run calculation as long as the output has not been deleted from the compute server. The only calculation states that cannot be "reconnected" are "created", "ready", and "imported". In fact it is possible to reconnect job monitoring to a calculation where job monitoring is currently running. In this case the current job monitoring processes are killed and new ones are started. This may be useful as a sanity check if something in the properties displayed in the Calculation Viewer looks odd while the output file seems reasonable.
Jobs started prior to this patch can themselves be reconnected when ECCE is available again. One caveat is that the completion state of the job will always be set incorrectly to "terminated". This is because the name of the file used to determine the calculation exit status has changed and only newly submitted jobs will recognize the new file. Simply use the "reset to state" function of the Calculation Manager to set the state to "completed" when it does change to "terminated". Newly submitted calculations after the patch and then reconnected will pick up the proper exit status.
The first time that the login window comes up only the "New" and "Cancel" buttons will be enabled. The "OK" button will be enabled on subsequent ECCE sessions once a passphrase has initially been set. The mouse pointer is automatically put into the login window and the passphrase input field has input focus without selecting it. Enter a passphrase and record it on paper if necessary until it is committed to memory. Hit the return key or the "New" button to accept the new passphrase. There will be a confirmation dialog when entering a new passphrase. You can reset your passphrase at will in the future even when you forget your old one. This is secure because all the encrypted information you set with a given passphrase is cleared when you enter a new passphrase so no passwords are compromised. Storing login passwords within ECCE is strictly a convenience that saves you from re-entering passwords numerous times during all the behind the scenes remote communication performed by ECCE. It is fully intended that you know or have recorded in some secure fashion the login passwords for all the individual compute server accounts you launch jobs to. Thus forgetting a passphrase only amounts to having to pick a new one and then re-configuring passwords either through the Job Launcher or the Machine Browser Configure Machine windows. It is not the intention of ECCE to let you set and forget all your compute server passwords.
When you enter a new passphrase the database login window comes up prompting for both a user name and a password. This is needed because the database server password is also encrypted using the new passphrase. At the time of the initial release of V1.5 all database server passwords on the machine avatar were reset to new values. In order to obtain your new database server password you must contact email@example.com. The reason for resetting these passwords is that with the old ECCE password encryption never prompting once the password is initially set it is likely that most users have forgotten their original database passwords. They are not forced to remain synchronized with other passwords such as AFS and DCE passwords and they have no expiration so they are likely original passwords dating potentially a couple years back. This is a convenient time to guarantee that new passwords are set. The user name for the database server will be your normal login name.
Once the correct database server password is entered the Gateway will come up. The database server password has been encrypted and saved using the passphrase supplied so the database login window will not come up again unless a new passphrase is set. You can either immediately re-configure the passwords for all your compute servers or you can do it as needed when you launch a job. All configuration information other than the password such as login name, remote shell, and calculation directory is retained when new passphrases are set. Thus it is only necessary to set the password in the Configure Machine window and hit save. One important shortcoming in the current design of ECCE is that applications started before a machine is configured will not be updated with the new configuration. Thus you won't be able to do operations like open a remote shell from the Calculation Manager on a newly configured machine until you close the Calculation Manager and restart it from the Gateway. Working within a single tool like the Launcher, configuring machines and then launching jobs, does not exhibit this problem.
The passphrase you enter starting ECCE is communicated between all the ECCE applications you start within a single session. It is only entered once when the Gateway is started with the "ecce" command. The passphrase itself is encyrypted using the same strong encryption algorithm used to encrypt your compute server passwords. This makes it extremely unlikely that a hacker will be able to grab your passphrase and use it to decrypt all your other passwords. The encryption used on the passphrase itself would have to be cracked before the other passwords could be decrypted. Further the encrypted passphrase itself only exists while the ECCE session is running and cannot be grabbed from any source while you do not have ECCE running. No clear text decrypted information is maintained in the memory of running ECCE applications.
For ECCE sessions after the first one where a new passphrase is set hitting the return key or the "OK" button will check that the passphrase is correct and is able to decrypt the encrypted passwords. If it is incorrect you can either re-enter it if you mistyped it or enter a new passphrase (clearing all previously configured passwords including the database server password).
The values used for default covalent radii are taken from the Sargent-Welch Scientific Company table of the elements, 1968 except for alkali elements, their ionic radius is used. Van der Waals radii may be established from gas-phase critical volumes or neighbor-neighbor interactions in crystals. Default van der Waals radii come from either A. Bondi, J. Phys. Chem. 1964, Vol 68, page 441, or the Sargent-Welch Scientific Company table of the elements, 1968. Some values for non-metals have been adjusted smaller to better emulate CPK radii. Elements for which values are unknown were assigned based on their position in the periodic table.
There are a couple of minor bugs associated with the clipboard:
- The new measures are not copied over to the clipboard views.
- On some occasions, you may need to rotate the clipboard object for the orbital transparency to display correctly.
RGB, GIF, JPEG, and TIFF image formats are supported. To make use of the the equipment in the EMSL Graphics and Visualiztion Lab, you should output RGB files with the size option set to "video".
Known bugs Top
- The same limitation with not being able to share project databases between Suns and SGIs that existed for V1.4.2 also exists for V1.5. This is due to an unaddressed bug in the ObjectStore database software. Refer to the V1.4.2 Release Notes for further details.
- Sizing problems with the Calculation Viewer property data panels originally reported in V1.4.2 persist in V1.5. To fix a property panel whose right-hand edge has been clipped grab and slightly move the small green square near the bottom of the vertical line that separates the property panels from the viewer area on the right.
What's fixed? Top
- (Sept. XX Patch)When running on an SGI, the Launcher or Machine Browser could crash due to a bug in the Machine Configuration dialog. The crash occured when changing the Machine Configuration password field. This bug has been fixed.
- (Sept. XX Patch)The Collaboratory project Electronic Lab Notebook integration described here broke when the latest version of the ELN software was released. A new version of the ELNWizard tool has been integrated and the Electronic Lab Notebook is used from within ECCE as it was before.
- (Sept. XX Patch)For large properties the perl parse scripts used to translate the data in the output file into a file format for the ECCE database can timeout. When the parse scripts take longer than 30 seconds to complete ECCE would timeout and the property being parsed would not be loaded into the database. This was discovered with molecular orbitals for very large chemical systems. The timeout period for parse scripts to complete has been bumped up to 5 minutes. This impacts both monitoring running jobs and importing finished calculations.
- (Sept. XX Patch)When trying to start a remote shell through functions such as "Tail -f on Output File" or "Shell in Calculation Run Directory" in the Calculation Manager, or "Open Shell" in the Machine Configuration Dialog among others, it would often fail when the remote server was on a different internet domain. This is due to the remote shell using the DISPLAY environment variable as set on the local machine which usually does not include the full domain name unless the user explicitly has set the variable before starting ECCE. The previous workaround was to exit from ECCE and set the DISPLAY variable with the full domain name (i.e, lionheart.emsl.pnl.gov:0.0 instead of lionheart:0.0) and then restart ECCE in the same window. The patch code now determines the domain name itself and starts the remote shells with a DISPLAY variable using this full name. Note that remote shells may still fail because of xhost access control lists not including the remote server. When running ECCE on the local machine ECCE will attempt to modify the xhost access control list itself. However when ECCE clients are running on a different machine than the machine X Windows are displayed on it is the responsibility of the user to use the xhost command prior to running remote shells.
- (Sept. XX Patch)A memory management bug was fixed related to the password text entry fields (those that echo asterisk characters instead of what is typed) that caused the Job Launcher to crash occasionally. Crashes would usually occur when selecting several machines from the machine list in succession where the password field would update for each of them and corrupt memory.
- (May 17 Patch)As a result of an mpp1 upgrade it was no longer possible to start remote xterm shells from the Calculation Manager, Launcher, and Machine Browser on that machine. The configuration of login nodes changed so that the xhost command needed that ECCE automatically performs to allow display to the local X server no longer worked. This has been fixed so the proper xhost command is issued. The new implementation also resolved a related problem on ecs1 so that it now displays remote xterms.
- (May 17 Patch)On SGI workstations only if the number of characters entered for the passphrase when first starting ECCE were fewer than the actual number of characters in the passphrase a deadlock would occur. The passphrase dialog would disappear but the Gateway would never start. This bug has been fixed.
- (Feb. 1 Patch)The Normal Modes property in the Calculation Viewer now automatically calculates sensible vector magnitudes when Vector Display is selected. The range of slider above the rendering area was reduced to a minimum value of 0.01 and a maximum value of 30 so that the meaningful range for scaling is no longer restricted to the lower limits of the slider. Whenever a new selection is made in the Normal Modes table, the magnitude of the vectors are recalculated such that the largest vector is 1 angstrom.
- (Feb. 1 Patch)Like the Calculation Editor Final Edit and Calculation Manager View Input File and View Output File features, the Manage Files View feature in the Calculation Manager now uses the $EDITOR environment variable. Previously the only editor that was used was vi regardless of the $EDITOR value. When viewing files on a remote compute server please keep in mind that whatever $EDITOR variable is set on the local machine will need to be in your path on the remote server to be found.
- (Feb. 1 Patch)Job submission files are now correctly generated for using multiple processors via the nproclinda directive for Gaussian 98 on mpp1 and nwtest. The latest upgrades on these machines had broken this functionality for a period of a couple weeks in late December.
- (Feb. 1 Patch)For machines using a queue system supporting a maximum memory limit, notably mulliken.emsl.pnl.gov, the field for the memory limit in the Job Launcher would sometimes be left empty. Whether or not the field was empty depended upon the sequence of operations in the Launcher prior to selecting the machine, e.g. mulliken. When hitting the Launch button the job submission would fail with a cryptic message about an invalid submit script. The memory limit field is now properly defaulted instead of being left empty eliminating the submission errors.
- (Feb. 1 Patch)Successful imports of PDB files through the Builder were previously being incorrectly interpreted as unsuccessful because the babel utility exits with non-zero values for success in certain situations. ECCE now recognizes the non-zero babel exit codes as successful.
- (Feb. 1 Patch)When ECCE was started it was possible for it to hang even before the passphrase prompt window was displayed. This was occuring in a small utility program that attempts to allocate UNIX "pseudo-terminals" used later by ECCE applications to manage remote communication. The allocation hung in a state where "pseudo-terminals" would remain allocated even after a ctrl-C was used to exit ECCE until certain processes were killed or the machine was rebooted. It was possible to restart ECCE and have it come up but most often attempts to launch jobs or other remote communication would fail. This had been observed only on one particularly troublesome Sun workstation.
- When building using nubs, the bond distance of created bonds is now the sum of the covalent radii. Previously all bonds were created with a length of 1.44 angstroms and presumed to be corrected by the force field clean operation. This may give some surprising results because it can look quite different from previous versions of ECCE. If you manually change a geometry, no attempt is made to recalculate the bond lengths based on the covalent radii.
- Bond generation now works correctly due to the improved values for the covalent radii. Note that you can control the bond generation by changing the values used for covalent radii (see "What's new?"). There is now a "Generate Bonds" menu item under the "Options" menu in the Builder.
- The CPK display now uses the van der Waals radii (VWR). All other display styles use the covalent radii (CR) scaled down by 50%. Other display details have been adjusted. Changes to the "Details..." windows under the "Display" menus for the Builder and Calculation Viewer are now retained as user preferences between ECCE sessions.
- Menubar menu items in the Builder and Calculation Viewer have been changed so they are now consistent. The "Display" menu has commands for modifying the display style and how the molecular system is rendered. The "View" menu has commands for turning on or off parts of the display like atom or bond labels. The "Options" menu has commands for editing colors, adjusting lighting, etc. The "Chemical System" menu is no longer needed in the Calculation Viewer. It is now possible to display bond lengths in the Calculation Viewer while animating a geometry trace or vibrational mode. An option to display or hide Hydrogen atoms is now available in the Calculation Viewer as well as the Builder.
- Bonding nubs now move with the surface of the atom they belong to when you switch display styles or change the sphere radius scale in the "Details..." windows under the "Display" menus in the Builder and Calculation Viewer.
- When performing single point calculations using Gaussian 9X, all results were reported using the geometry supplied by the Builder, even though Gaussian frequently transforms these coordinates at least once. Results are calculated in the transformed coordinates and may be meaningless if displayed in the Calculation Viewer using the original geometry. This applies particularly to dipole moments and molecular orbitals. This is not a problem for geometry optimizations. ECCE now updates the geometry for single point calculations if a transformed geometry is found.
- The DFT correlation and exchange functionals were being specified in an incorrect way in the input decks for NWChem if a non-default value for either the exchange or the correlation functional was specified in the "Theory Details" window of the Calculation Editor. No problems occured if a non-default combined functional was requested. The problem was that ECCE was requesting the exchange and correlation functionals using separate XC cards in the DFT field for NWChem, which does not work. The functionals are now requested on a single XC card in the input deck.
- The restriction of only being able to run calculations in a directory on a local disk has been removed. It is now possible to specify an NFS, AFS, or DFS calculation and scratch directory. Before this could only be done by setting the ECCE_ALLOW_ANY_FS environment variable. Note that is still highly recommended that calculations be run on local file systems because the input/output speed for shared file systems is remarkably bad.
- Calculation imports through the Calculation Manager now have better recognition and warning messages for running out of disk space while importing. These were often reported as "unknown" failures in the past.
- Calculation imports through the Calculation Manager now have better end-of-file detection. Previously an end-of-file before the end of a parsed property was reached resulted in an infinite loop where the Calculation Manager would have to be killed.
- File transfers performed when launching a calculation and copying results back upon completion have more reliable error detection. Previously errors due to improper file permissions for copying and lack of disk space would go undetected and the copy would be reported as successful.
- User preferences for atom colors and radii set through the Periodic Table application or through the periodic table window in the Builder are now saved immediately. Prior to V1.5 an extra step was required to save any changes and a "dirty star" save button was displayed whenever a change was made. This change makes the interface less confusing. Both the color and radii values can easily be reset to default values.
- Distance, angle, and torsion measures displayed in the Builder are now updated after performing a Force Field clean. Previously, they did not correctly update as the location of their associated atoms changed.
- Error detection and reporting while importing molecules has been improved for the cases where the babel utility is used for the conversion (everything except XYZ and MVM).
- Coordinates in the input file are now dumped out with 10 digit precision to improve the chances of correct symmetry detection.
- The color spectrum in the mulliken charge display of the Calculation Viewer now correctly handles the case where the minimum and maximum charge values are the same. An artifical spectrum range is created in this case. This was generally only a problem with diatomics and resulted in a core dump of the Calculation Viewer on SGIs.
- Both NWChem and Gaussian 9X jobs are now "terminated" with SIGTERM. Previously, NWChem jobs were killed with SIGINT to avoid shared memory problems when using parallel.
- NWChem ecce.out files with the coordinates expressed in bohr are now correctly converted to angstroms.
- Attempts to use basis sets containing i functions with NWChem are now trapped and prevented. Support for i functions has been disabled in NWChem.
- When using the ECCE_NWCHEM environment variable to control which version of NWChem is used, a bug in the generated nwchem.p file has been fixed. The executable path was not correctly substituted into the file.
- Previously it was not possible to select a project database formatted for a previous release of ECCE for deletion through the Calculation Manager "Open Database" window because a "Migrate Database" dialog pops up immediately. Dismissing the dialog also unselected the database in the list. This resulted in a "catch 22" situation where reselecting it would pop the "Migrate Database" dialog up again. Now the database will remain selected after dismissing the dialog and hitting the "Delete" button will remove it.