Transitioning to the IRIX(TM) Operating System

A guide for system administrators

Bill Middlecamp
Silicon Graphics, Inc.
655F Lone Oak Drive
Eagan, MN 55121

wjm@sgi.com

ABSTRACT:
The roadmaps presented by Silicon Graphics® show that the successors to traditional Cray Research® architectures will use the IRIX operating system. However, the UNICOS® and UNICOS/mk(TM) operating systems offer many specialized tools for the management of supercomputers. Therefore, Silicon Graphics mobilized immediately after the merger to carry forward these capabilities to the IRIX system. Now, Silicon Graphics is collecting information into a document for system administrators to help them learn the IRIX system and plan for its place in their future supercomputing resources. This presentation offers a preview of the document, which will be published by the end of 1998, and asks for feedback regarding the document's structure and scope.
KEYWORDS:
Operating Systems, System Administration



Copyright © 1998. Silicon Graphics Company. All rights reserved.

Introduction

The IRIX system is the long-term direction for the Silicon Graphics technical high-end server products. The UNICOS and UNICOS/mk systems continue to be enhanced for the CRAY T3E(TM), CRAY T90(TM) and CRAY SV1(TM) products.

The IRIX, UNICOS and UNICOS/mk systems are all based on the UNIX system, so their similarities are greater than their differences. The transition to the IRIX system is far easier than was the transition from the COS(TM) system to UNICOS system.

To make the transition as easy as possible, Silicon Graphics is writing a guide for system administrators: Transitioning to the IRIX Operating System, publication SN-2241. The purpose of this document is to do the following:

The documentation project began at the start of this year, and progress has accelerated since the completion and announcement of the Silicon Graphics product roadmap, but many of the projects described in the document have been underway since the early days of the Silicon Graphics/Cray merger. Much of the information for SN-2241 has been collected and is being reviewed for accuracy at this time.

This CUG paper offers a preview of the document, which will be published before the end of 1998. Our objective is to obtain customer feedback on the document scope and topics.

Structure of SN-2241

Each chapter contains the following:

Note: In all cases, the information is subject to change. We expect the system to evolve as we discuss plans with you and as we respond to new developments.

Topics Discussed

The following outline gives a preview of the topics discussed in the document. Please verify that all the administrative topics that are important to you are covered here:

  1. File System Planning
  2. Configuration
  3. Cellular IRIX System
  4. Startup and Shutdown Procedures
  5. Basic Administration
  6. Networking Facilities
  7. Job Management
  8. Resource Administration
  9. User Database
  10. Tape Subsystem Administration
  11. Cray Data Migration Facility
  12. System Dumps and Debugging
  13. Security
  14. Software release management

Sample Chapter

The level of detail varies among chapters. The following example is a typical chapter:

Chapter 11. User Database (UDB)

The UNICOS and UNICOS/mk user database (UDB) contains an entry for each user allowed to log into and run jobs on your system. The UDB, which replaces the traditional /etc/passwd authorization file, allows fast access to an individual user's information.

UDB functionality will be an optional component on future IRIX systems. From the UDB perspective, the transition from UNICOS and UNICOS/mk systems to the IRIX system can be broken into the following categories:

In UNICOS and UNICOS/mk systems, all UDB information was user based. This will also be true in the IRIX system, although the specific items that are supported will be different.

Note: In the IRIX implementation, the UDB is not involved in authentication. Instead, the UDB will contain only user limits information.

The IRIX UDB implementation is layered upon the Pluggable Authentication Modules (PAMs) feature, which provides the means for enabling the optional UDB feature. The PAM infrastructure is defined by the LINUX organization. The PAM feature will be available in a future IRIX release. For information about associated process limits, see Section 10.2.1.

11.1 Functional Differences

The IRIX implementation of the UDB will contain only limits and nice values; other functionality that is provided by the UDB in UNICOS and UNICOS/mk systems is provided by other means in the IRIX system. Table 4 shows where the UNICOS and UNICOS/mk UDB capabilities are provided in the IRIX implementation.
Table 4. UDB Capabilities in the IRIX Implementation
Function IRIX Implementation
Process and job limits (including separate values for batch and interactive Future UDB implementation
nice values Future UDB implementation
Identification and authentication information Current IRIX authentication method
Fair Share data Current Share II product (layered feature)
Account IDs Current project IDs are equivalent
Group IDs Current group file
Last login time (Does not exist)
Home directory Current password file or NIS
Password aging information Current IRIX implementation
Security information Current IRIX implementation
Perm bits Current IRIX privilege mechanism provides equivalent functionality

The specific process limits that will be provided in the IRIX UDB implementation are those limits that are already supported in IRIX system. The big difference is that they will be made user-based whereas IRIX limits are currently system-based. The new job limits that will be supported are as follows:

Table 5 shows a list of the items that will be supported in the IRIX UDB. Some IRIX symbol names have not yet been defined.

Note: UNICOS/mk specific limits do not have counterparts on the IRIX system, so those fields will not be carried over.

 
Table 5. IRIX UDB Supported Items
Resource Process/Job UNICOS & UNICOS/mk IRIX
CPU time limit Both ue_pcpulim, ue_jcpulim RLIMIT_CPU
File size Both ue_pfilelim, ud_jfilelim RLIMIT_FSIZE
Memory a process program can allocate Both (none) RLIMIT_DATA
Stack size Both (none) RLIMIT_STACK
Core file size Both ue_pcorelim RLIMIT_CORE
Number of open file descriptors Both e_pfdlimit RLIMIT_NOFILE
Virtual memory size Both (none) RLIMIT_VMEM
Physical memory size Both ue_pmemlim, ue_jmemlim RLIMIT_RSS
Number of threads Both (none) RLIMIT_PTHREAD
Number of processes Job ue_jproclim TBD
Job tape unit limit Job ue_jtapelim TBD
nice number Process ue_nice TBD
Disk space protected from archiving Process ue_archlim TBD
Archiving media selector Process ue_archmed TBD
 

Note: In some cases, there are both process and job values. These fields will be implemented in the IRIX system.

Table 6 shows which related interfaces will be provided in future IRIX systems.
Table 6. UDB Related Interfaces
Interface Description Future IRIX Availability
udbgen(8) Command line program that allows a root user to add and delete a user record Yes
udbsee(1) Command line program that allows users to see their own records and allows a root user to see secure information (such as passwords and security settings) Yes
udbchain(8) Command line diagnostic that verifies the integrity of the data files Yes
libudb.a(3C) Library used by udbgen, udbsee, and udbchain to interface to the index and data files. Yes
nu Command line program that adds and deletes user accounts. No
xadmin Graphical user interface that adds, deletes, and modifies user accounts. No

11.2 Procedural Differences

A UDB is generated on UNICOS and UNICOS/mk systems from either the output of a previous UDB or from an existing password file. The same will be true on IRIX systems.

The ASCII output from udbsee -a on a UNICOS or UNICOS/mk system can be used as input to udbgen on IRIX systems. In this case, the IRIX udbgen command will take what is relevant from the original output and assign default values to the remaining supported items.

In the case of using the password file to generate the UDB, all users represented in the password file will be added to the UDB, but they will all have the default values for all supported items.

Another difference in the IRIX version of udbgen is the implementation of defaults. On UNICOS and UNICOS/mk systems, the defaults were compiled in and they were stored differently from other user records. On the IRIX system, they will be read in at runtime from a special default record file and they are stored as specially marked user records. In UNICOS, UNICOS/mk, and IRIX systems, the defaults can be modified to fit specific needs. This change was made to improve the maintainability of udbgen by simplifying and generalizing its structure. A fallout from this is that udbgen no longer has special knowledge of the values assigned to the default record. To accommodate this, an ASCII version of the default record will be released with the UDB. However, this record must be read in prior to building a new UDB. If the default record is not read in before an attempt to add users, udbgen will fail. Thus the procedure is to read in the default record before using the password file or udbsee output to add users.

Distribution of SN-2241

SN-2241 will be offered online (in Postscript and PDF formats) at the CRInform web site so that it can be a living document, evolving in response to your requirements. Its availability will be announced in the Cray Research Service Bulletin (CRSB). Feedback about SN-2241 will be collected by email. For now, you may send your comments about this preview to api@cray.com, the email alias for responding to the SuperComputing API.

Request for Early Feedback

We expect to publish SN-2241 to all customers before the end of 1998. However, we would like to get some early feedback about the document from a small number of customers. Our expectations for these reviewers are that they will agree to the following:

If you are interested in participating in the review process, please submit your name and email address to api@cray.com. We will choose a few reviewers when the document is ready.

Summary

Transitioning to the IRIX Operating System, publication SN-2241, is a new document under development at Silicon Graphics that will help administrators understand and plan for their migration from the UNICOS or UNICOS/mk systems. It provides definitions of key terminology and high-level descriptions of relevant functionality. It also serves as a focal point for discussing administrative requirements and evolving plans. The document will be made available online before the end of 1998, and feedback will be collected through an email alias.

Table of Contents | Author Index | CUG Home Page | Home