Introduction: 

This discussion paper is intended to give an overview of the current Waterloo Polaris system (referred to as Phase I) and future directions EC staff would like to take in developing a more robust, more easily supported environment referred to in this document as Waterloo Polaris Phase II.

 Waterloo Polaris Phase 1:

 Waterloo Polaris Phase I was developed as a Win32 successor to the 16bit Watstar and as a foundation for future enhancements to our computing environment. Waterloo Polaris Phase 1 provided the familiar Watstar user features in a new Win32 environment.

 Phase I also adopted some significant changes which will act as a springboard for us to move to the use of open technologies as the building blocks for file, application, e-mail, web and other serving functions. This will lead to the elimination of the Watstar server over time:

 Unix or any other open server could become the Email storage host, this replaced the proprietary Email delivery technologies used in the past

• Any IP routing technology could be used, eliminating past practices of a particular 3^rd party router product. Today the Watstar server adds security as an IP firewall, but in the future we hope vendors like Cisco will be able to assist.
• Industry standard switched Fast Ethernet became the norm for faculty backbones (and soon the University backbone), eliminating the older single-vendor proNET technology which was used in the past. Fast Ethernet workgroups are also becoming common.
• Any SMB/CIFS file server such as Unix, NT, Network Appliances, etc. could be used for Web and file storage.

 Phase I was the most difficult and expensive step which required that we upgrade many workstations, all the software, the Email subsystem, the networking technology and upgrading all the servers.

 From the systems software side, Phase I was also the most complicated because we had to solve new Windows problems and create and perfect a lot of new technologies. Win95 comes out of the box knowing how to talk to NT and Netware servers, we had to make it talk Watstar, and make it respect Watstar’s legacy features which are highly desired but unavailable from Microsoft. We also had to make Win95 talk to Unix and NT systems better than it does by default.

 In winter 1998, there will be some incremental additions to improve multi-platform and Internet issues on Phase I. Most involve minor client software enhancements, and these changes will be distributed automatically to all Waterloo Polaris stations.

 increased Email security to reduce tampered messages

• automatic password synchronization feature for Email and SMB/CIFS servers
• automatic P: SMB/CIFS drives if selected by system manager
• automatic SMB/CIFS printers if selected by system manager

One of the ambitious design goals of Waterloo Polaris Phase I was backward compatibility with Watstar. However, Waterloo Polaris is NOT permanently tied to the Watstar server. That connection is only temporary to facilitate the transition - Waterloo Polaris Phase I currently relies on Watstar and established subsystems to provide continuity. In order to move to a more open systems architecture, we need to abandon the Watstar server.

Phase II

The most important aspect about Phase II is that we have a functional Phase I system in place, giving us the freedom to implement Phase II in a modular fashion, at our convenience, rather than in a rushed fashion.

With Phase I in place, Phase II is more of an architectural shift than a user shift. In fact, most typical users will be unaware of the next changes. But users who somehow want to push the envelope (e.g. village users, office machines, laptops, etc.) will all be enthused. We also expect the faculty computing offices and IST to be pleased as this will improve their ability to manage and support the resulting system.

The Client OS

The client OS supported by Waterloo Polaris Phase II will be the standard, or preferred, client OS, currently Windows 95, and a ‘light’, or secondary client which will allow other client OS, and a core, but reduced set of features as compared to the primary client OS.

 1. The Primary Client OS

 The next phase of Waterloo Polaris will have a preferred client OS which will likely remain Windows 95 for the present time. All student lab machines and any office machines hoping to have automatic updates and no maintenance should stick with the preferred client OS, and they will continue to have the most fully integrated client system, and the most fully supported one too. Win95 remains the best choice for now. Waterloo Polaris II for these preferred clients will not likely require a re-install, the few changes should be distributable using Waterloo Polaris itself

 2. The Secondary Client OS - the ‘light’ Client

 Other operating systems like NT and even personal workstation running most popular OS’s (desktops or laptops running Win95, NT Workstation, Linux, etc.) in locations like offices or the villages will be able to gain access to Waterloo Polaris II by using a ‘light’ client, which will provide only core Waterloo Polaris features:

• login authentication
• file serving
• data backup
• distributed printing
• web services
• internet access

 The effect of incorporating these systems into the campus PC network will allow a greater degree of access to the network by a larger set of more varied clients - thus providing a higher degree of integration than currently exists on campus. It will hopefully slow the need for additional computing stations and thus save money and space.

 The Server OS

 The primary goal in this second phase of Waterloo Polaris is to support open protocols and thus open server technologies. The philosophy toward open systems is easy - if a given server product speaks the right protocols, has the right capabilities, and is easily configured and managed, then there should be only minor dependencies for each supported brand.

An example is our present pop mail system, where only a small C-shell script exists to interface our generic account creation software between the various brands of Unix used to implement the mail servers. Upgrading to a more powerful server would be simple, and even supporting new brands of Unix is not a major issue.

An immediate result from this fact is reduced vendor reliance and improved campus stability. The life cycle should be measured in terms of years, not months, and we encourage vendor competition by forcing vendors to remain compatible with open standards.

Phase II Server Specifics

The general role of today’s Watstar server will be broken into several categories. Depending on the situation for a given faculty, some of these tasks may be grouped onto the same server:

• authentication server - this would authenticate user logins and match the user with his appropriate disk server, application servers, access rights, Email server, etc.
• user file server - this would be a specialized box dedicated to serving user files. It would have an appropriate backup strategy and RAID disks for reliability. This may be accomplished with one powerful server for some faculties, or a server farm for others. This beast would require user quotas.
• user application file servers - we would likely have a battery of servers (possibly doing other things too) which would do the simple but demanding task of supplying application files not already available on the workstation. Currently we have one such server per lab, or set of offices. User quotas not required here because no user files would be stored on these servers.
• Email server - already in existence
• Printer Server - an upgrade path for those using Watstar print servers today.

 Other concerns, like DNS and News servers are already covered by appropriate Unix machinery.

 Aside: The file and application server products will have to be well configured and used intelligently.

• The hardware specs to do this will be quite powerful. Anyone thinking such a beast was available before 1997 would be mistaken. We are still having difficulties finding them as we approach 1998. Unix system administrators have seen their systems stressed just by our user population’s Email needs, the requirements for file access will be greater.
• it could not be a general purpose computing system with user terminal logins. Session logins significantly impact resources and would detract from its performance and reliability.
• the Waterloo Polaris model is needed to make this possible (very different from the Watstar model), Waterloo Polaris uses locally installed applications. 95 keeps hundreds of system files open while it runs, so server based applications would present significant and unnecessary server load. This was not a problem on Watstar server because they, like NFS, did not maintain file states.
• by centralizing user files to one or two locations per faculty, we can afford to use RAID disk technologies to reduce risks caused by disk hardware failures. (Math’s NetApp does this already)
• We plan to perform some tests to show actual performance and resiliency under a simultaneous load of 30 and 60 machines before even considering a particular brand. This is a preliminary test because real classroom situations include more than one hundred simultaneous logins.

 Achieving integration with Open Systems

 Waterloo Polaris will adopt standards based mechanisms for logins (including directory services with LDAP), file service (CIFS/SMB), etc. in addition to our existing support of standards like POP/IMAP. On the surface these goals sound similar to NT 5.0's goals, but the differences are significant under close inspection - Waterloo Polaris will be designed for and committed to a truly open multi-vendor system with a focus on stability rather than a series of upgrade sales. We can concentrate solely on our needs and have the freedom to select and include proven server technologies as necessary. We easily build on what Microsoft offers, but massage it so the result is more open and more timely, but with no loss of functionality.

 Benefits of Open Technology:

 The benefits of an open technology for the file server are tremendous:

• Hoover or other open backup solutions would exist, taking the backup/restore load off of EC and further standardizing campus backups
• user authentication could be accomplished with Kerberos, timed passwords, or many other features which are presently unavailable on Waterloo Polaris.
• some open servers allow us to choose products with required features like user storage quotas and good NFS access for Unix users, whereas an NT-only approach lacks these, and its architecture prevents robust implementations.
• SMB systems allow access to residences, and home access once higher speed media appear (like RogersWave)
• open servers usually have built-in support for FTP serving, so that too would improve home access
• other support mechanisms will then exist besides relying on EC staff for Watstar server answers
• some administration operations are more easily done in a Unix-styled environment. For example, populating user accounts with the initial data files is easier in Unix, changing the disk quota for a user can be done faster, cron’d operations exist, and SMB session login/logout can prepare and postfix the account for a user if required.
• the resulting servers would likely have long term upgrade paths and options
• by selecting a scalable server system, we reduce the number of userid-bearing servers to one or two per faculty, and can afford to invest in RAID disks, better UPS devices, and other improvements.

 The Path from Watstar to Waterloo Polaris II

 Table 1 lists enhancements which have been performed specifically for Waterloo Polaris, but one can see that they are compatible with a non-Watstar future. Also listed is a suggested Phase II configuration for comparison.

 Table I - Features of the Transition Path.

Phase II Benefits

 The goal of Phase II includes selecting an appropriate server technology (NT, Unix, Network Appliance, etc.). Then Waterloo Polaris becomes simply the campus architecture and the workstation drivers which work with that server and the built-in Microsoft networking on the client.

 Security:

•  stations will be more secure when we no longer have to support DOS in real mode
• we gain the option of using TCP/IP to authenticate the user, an actual requirement if we adopt a campus authentication server which our plan is well positioned to do

 Performance

•  faster Logins and Logoffs when we avoid having to do the real mode client login and logoff, right now we pay for the compatibility maintained with Watstar (because real mode logons have to load protected mode as the user waits).
• improved network disk performance because we can use a server with a lazy write strategy, Watstar never had one.
• availability of scalable server technologies
• ability to use latest NDIS 3 and NDIS 4 drivers on the client (currently forced to NDIS 2 due to real mode startup)
• faster/easier to populate user files before the user’s first login because we can ask the server to do this for us.

 Support Benefits

• move towards faculty level fileservers which will facilitate economies of scale.
• phase-out of real mode network drivers - backward compatibility with DOS forced EC to use older NDIS 2, or DOS ODI technology. Once real-mode is phased out, we will be able to use more common (more current) protected mode NDIS 3 or 4 drivers.
• backups and restores can be automated through Hoover, not necessarily Eng. Computing
• system administrators get benefit of external support groups and 3^rd party tools
• decreased reliance on Engineering Computing staff for Watstar-specific information.
• continue to use Waterloo Polaris’s protection measures for the local machine, e.g.. disk locking

• account administration can be more automated with more intelligent servers
• long filename issues go away because we will have no real mode and its short name problems
• printing can be improved to be better than default Microsoft, or current Waterloo Polaris mechanisms
• improved CRAS integration will eventually eliminate the need for the current Watstar printer accounting

 User Benefits

•  the transition from Phase I to Phase II will not involve significant changes to how users see the system.

• Improved login speed - currently we carry out many 'convoluted' actions to support to hybrid Watstar/Waterloo Polaris Phase I environment. Skipping the dual-mode will allow a faster login strategy
• improved home and multiplatform access to files
• the ability for casual workstations -someone in their office running NT or 95 to access their network account without the machine committing to a full Waterloo Polaris installation.
• home and village access to files
• ability to support workgroups and shared projects better than a Watstar-only method. Waterloo Polaris already has this, but few taking advantage of it yet because few administrators are issuing SMB accounts
• reduced requirements for standardization of specialty machines

 Challenges

 There are some challenges associated with the Waterloo Polaris II approach:

•  This is an enormous system with enormous requirements. A typical day can include 17,000 Waterloo Polaris/Watstar user logons, the user authentication server would have to handle those and the logout requests. The number of file transactions would be orders of magnitude higher. It’s a formidable task to keep up with such load.

• Watstar servers are among the most reliable on campus. System administrators are understandably hesitant to abandon the Watstar technology and have raised questions about the reliability of successor technologies. We plan to conduct some tests in Science and Engineering, and to monitor the experience of Hooke in Math.
• The first truly usable revision of the LDAP protocol (v3) is not finalized yet. Previous versions are too insecure or incapable for our needs. Due to the release timing, it is likely that early releases of products like NT 5 will have LDAP interoperability problems. We will have to bridge such problems as we likewise overcome other problems.
• SMB/CIFS servers are stateful (they keep state information on all open files). This creates a lot of server load which the essentially stateless Watstar server was able to avoid. So faculties with lots of user stations will need servers capable of such high loads during peak periods. We may solve this with a server farm, but that also introduces other complexities and cost.
• Since the server file handles and file locks are limited resources, we will be open to problems or attacks of resource starvation. This was not possible under Watstar.
• Given the student residences and the possibility of Roger’s Wave and Bell’s ADSL, server load may be increased by home users who can remain connected at all times.

 Conclusions:

At this point Waterloo Polaris Phase II appears to be an attractive alternative to the current proprietary Watstar. It proposes to address many of the current concerns with Watstar and Waterloo Polaris Phase I, but clearly it needs further investigation and study.

Appendix 1: Some Server Platforms Notes

 Network Appliance

Unix

NT