Release Notes 3 0 1
This document describes changes for SRB 3.0.1, released on December 19, 2003.
This is the follow-on release of our latest major new version, SRB 3, the Federated MCAT, also known as the zoneSRB and Peer-to-Peer Federation. Our previous release, SRB 3.0, October 1, 2003, was Beta.
The new features include improved interoperability between SRB version 2 and SRB version 3, new features needed to improve performance, improve zone management, organize collections, and fix some bugs.
SRB 3 servers can now be built compliant to SRB 2 MCAT and SRB 2 clients. Run configure and then comment out the FED_MCAT = 1 in mk.config, and the SRB will have none of the Zone features but all of the other SRB 3 features and bug fixes. In this mode, you can run with an SRB 2 MCAT without upgrading it. The SRB 3.0.1 Server code will recognize SRB 2.x clients and use the old protocol, so you can update some or all of the clients later. All servers must be upgraded to SRB 3.0.1. In this mode, 2.x inQ will also work.
MySRB now operates with SRB 3.x and can display Zone information.
The SRB Java API, Jargon, has been updated to operate with SRB 3 (i.e. the new protocol).
UNIX_NOCHK and HPSS_NOCHK resource types were added which can be used when one wants the driver to not check for the existence of ?USER.?RESOURCE in the path. To activate this feature, use the "Add New Token" menu of the Java Admin GUI to add two ResourceTypes - "UNIX_NOCHK file system" and "HPSS_NOCHK file system".
The install script, install3.pl, has been updated to support the latest Postgres (7.4) and Postgres-odbc (07.03.0200), and to increase the stack size on Mac OS X (needed for this release). Some bugs, when run under Solaris, have been fixed.
There is a new Unix I/O library, unixIO, which that emulates Unix I/O calls, converting them to SRB calls. This is similar to srbio, but the application directly calls either SRB or Unix functions. unixIO has some different emulations for SRB features.
2) Improve performance
Sput/Sget can now do bulk upload/download (previously bulk data transfer was only available via Sbload and Sbunload). Also, bulk operations can be done on regular files (non-Containers). Use -b to request bulk operations. Sget -b will get from non-Containers, Containers, either, or both. Currently Sput -b will put to non-Containers (use Sbload to put to Containers). Compared to regular files, bulk transfers of large sets of files are often 5 to 10 times faster. Bulk transfers to/from Containers are often 20 to 50 times faster.
The general query protocol has been modified to greatly reduce the size of messages sent across the network. Users should see some improvement in latency in queries such as Sls, especially if the distance between the client and the MCAT server is large.
Changes have been made to reuse sockets quickly when linking with Oracle library (add SO_REUSEADDR option in srbSetsockopt()). This is normally default, but needs to be reset when linking with Oracle.
3) Improve zone management
The Java admin tool has improved Zone management functions. When creating a Zone, it now lists various options to choose from as is done in other panels. Also, new panels have been added to display resources information, in both summary and detail form.
The zonesync.pl script has been enhanced. Run 'zonesync.pl -h' for additional information.
4) Improve collection organization and management
Links are now supported. These are like Unix softlinks, where one can create another logical name to point to an existing data object. The new command is Sln (like Unix ln), and Sls will list these links. The limit to the number of characters allowed in a SRB path name has been increased from 200 characters to 500 characters. GUID (Global Unique IDentifier) support has been added. A GUID will be generated for each data object stored (if so configured by the administrator). GUIDs can be modified via a SputD command and viewed via SgetD. GUIDs will be useful when interfacing to other Grid software systems. A parameter, STOR_VAULT_MODE, was added in runsrb which allows the access mode of files created in the vault to be changed to values different than the default value of 0600. Srm has been modified so that it can successfully remove collections even if the container and in-container files are in the same directory. (By default, containers are created in /container but if users create containers in a collection, Srm will now handle it.) The Sbunload interface was significantly changed as per user request. See the Sman page for a description. Also note that Sput -b can be now be used for Sbunload operations. 5) Bug fixes and documentation improvements A number of smaller bug fixes and enhancements have also been made. Updated documentation includes changes and additions to faq.html, README.MCAT.INSTALL, and install3.pl.