SCOREHOSTS.DB(5) | SCOREHOSTS.DB(5) |
A record is terminated by a newline (\n) character. The record name is the first blank-separated token followed by one or more attribute/value pairs. Each attribute name and value pair is also separated by blank(s) or tab(s). The equal (=) character, immediately after the attribute name, separates attribute name and value tokens. Lists of values are separated by comma (,). There is no way to include a blank character in a record name, an attribute name, or a value.
Here is an example of a record:
a b=1,2,3 c=trueThis describes a record "a" having two attributes "b" and "c". The values of attribute "b" are "1", "2", and "3". The "c" attribute has only one value of "true". There can be multiple attributes with the same name in a record, and in this case, the values list is treated as a concatenation of the attribute values.
A value can also be path to a file. When there is a record like:
dwarf bib:file=/home/dwarf/bibliography.txtIn this case, the attribute name must be suffixed by :file. If a client program running on a remote host tries to read the file located on the server host where the scoreboard program is running, the scoreboard program and its client library together make a cache file on the client host. The cached file is located in the directory /var/scored/scoreboard, and the cached file name is encoded to identify the original file.
The database file is firstly preprocessed by the cpp(1) UNIX program when the database file is analyzed by the scoreboard program. Thus users can include any other database file (using #include) or define (using #define) any macros in the database file. INSTALL_DIR is a predefined symbol and set to the same directory where scoreboard was installed. Consult your system administrator for the exact location. All parentheses ("(" and ")") are ignored, however, unmatched parentheses are warned. Usually parentheses are used to separate macros with other symbols. Here is an example.
troll datafile=(INSTALL_DIR)/troll.txtWhen a client tries to open and read this file, the original file is located in the INSTALL_DIR directory, and INSTALL_DIR is defined when scoreboard was made.
* Attributes used by the SCore Cluster System Software
The Cluster Database file must hold all cluster configuration information required by the SCore programs. There must be two types of records, host records and network records.
The record name of a host record must be exactly the same name as the hostname of the host. The order of host records affects the order of node numbers in a group. The first host of a host group, from the beginning of the database file, will have node number zero (0).
Required attributes depend on which SCore programs will be run.
Program name | Attribute name |
msgbserv | group msgbserv |
scout | group msgbserv [spare] |
scored | network cpugen speed smp [watchfs] |
The attributes are as follows:
hostname
(1) command.
The port number, if any, must be an even number.
Each network record must have one attribute, type. The value of the type attribute must be the same as the PM device name (e.g. type=myrinet). The order of network records is arbitrary. All other attributes vary from device to device. The name of this record might be specified scrun command.
PM device | Attributes |
myrinet | -firmware:file -config:file |
ethernet | -config:file |
shmem | -node |
agent | -unit -agent -config:file |
The attributes are as follows:
On PM/Agent, this attribute is optional and only needed if
there are nodes which have to run multiple agents.
For example, if your site has multiple IP networks which are
not reachable from each other, and if there are nodes which
have multiple IP network interfaces connected to both of the
IP networks, you may specify multiple PM/Agent/UDP
configurations for each interface by specifying different unit
numbers as follows:
udp-100base type=agent -agent=pmaudp \ -unit=0 -config:file=(INSTALL_DIR)/etc/udp-100base.conf udp-1000base type=agent -agent=pmaudp \ -unit=1 -config:file=(INSTALL_DIR)/etc/udp-1000base.conf