Versions Compared

Old Version 4

changes.mady.by.user Alex T Prengel

Saved on

compared with

New Version 5

changes.mady.by.user William D Cattey

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: Deleted alexp's comments from the public facing version. They're still in the old version which you can get to off the info tab.

...

No Format
nopaneltrue
A. Network
B. Kerberos/Certificates
C. Printing
D. 3rd Party Software
alexp: to date, most IS&T-supplied Linux Third Party applications have been delivered via AFS lockers on Athena,
and are also accessible on any machine running the AFS client. Key components listed here by letter are either part
of the OS, or available (mostly) at no cost through the IS&T Linux download site. A limited number of
proprietary Third Party apps are also available from the IS&T VSLS site (most at some cost). Various
departments purchase their own apps, which may or may not be on IS&T's radar. A large subset of applications
are free/Open Source; many are available through the OS packaging system, though some are not. Currently,
Ubuntu seems to provide the broadest range of packages. On Athena, many that were previously in lockers are
being shifted to OS (Ubuntu) packages, to the extent that this is feasible. This is a very efficient way of
installing software: very easy to do, and very low marginal cost to add additional applications in a centrally
controlled environment like Athena 10 compared with, for example, building an application from source in a locker.
The downside is that applications can be out of date, sometimes badly.alexp: For commercial software we need to pay for by license, by far the most efficient way to distribute a
relatively small number of licenses to a large number of users is to use floating licenses controlled by
a license server. It's often surprising how few licenses can serve a large number of users; one downside is
occasional peak loads using up all the licenses (day before a big project is due).
No Format
nopaneltrue

E. Email/Calendaring
F. Instant Messaging
G. Web Browser
H. Security/Software Update
I. VPN
J. Backup
K. Filesystem/AFS
L. Remote access configuration: Athena 10 proposes to eliminate this.
alexp: many of the components we supply above at the IS&T Linux download site require custom tweaking for MIT.
This can also be a complication in supplying other software via OS packages in some cases.

Classes of Service

Based on who owns/manages the system:

  • IS&T Owned and Managed
    alexp: this makes it relatively easy to centrally distribute software. machines are generally in a common, known hardware and software state, and installing it for one machine (usually) ensures that it will work on all. A big advantage of doing it in network file systems is the ability to roll out updates quickly (sometimes on a few hours' notice if urgent).
  • Managed by Department Lab or Center alexp: this can make it harder for IS&T to deliver/manage software for them, particularly if they don't have AFS. Often we supply software (or they acquire it independently) and we delegate installation/maintenance to them. It's harder to tailor preconfigured software for every possible circumstance.
  • Managed by individual owner.
    alexp: this makes it even harder for us to preinstall/centrally configure software for them. For example, a common situation with non-Athena users accessing locker software via OpenAFS is that various required dynamic libraries are missing, which need to be tracked down and installed. Applications available through the OS packaging system are easy- users can just be advised which to install. In many cases with commercial software, the best we can do is supply the installer and installation instructions.

Use Cases

  • General Purpose Computing
    • Connectivity/Communication
    • Document Preparation
      alexp: I'd roughly estimate that Email/Web browser/Office Suite/pdf viewer comprise the vast majority of software launches on Athena, probably 70-80% or so, and that comparable numbers apply to Linux generally.
  • Courseware i.e. Applications tailored for use in teaching.
      alexp: The majority of this is in course lockers/departmental servers maintained by others and off my radar. In former days when I worked closely with Faculty Liaisons focused on Athena, I had more awareness of educational computing in departments/courses. As Academic Computing shifted their attention elsewhere and moved out of IS&T, this connection has been significantly weaker for me. I've suggested in the ACCORD group that we should look for way to strengthen it.
      • Often consistency, e.g. the same app. version for all students in a class is important.
        alexp: It's important to not change default versions of an application during a running semester. With central management it's easy to do this. Conversely it's sometimes necessary to have multiple versions of an application around; the AFS locker system I've been using makes it easy to do this, and it can be hard or impossible to do this with OS packages.
    • Research i.e. Applications tailored to scientific use.
      alexp: many specialized applications on Athena are of this nature- used by small numbers of users but of critical importance to them. Support can be difficult as these are often highly technical. Often we can provide at most installation support, i.e. making sure the application is running correctly or providing an installer.
    • Special Purpose Computing
      • Distributed Computing alexp: "seti@home-type" distributed computing has not been supported by IS&T (on Athena anyway) due to a variety of resource/technical issues that make it impractical; this is likely to be an ongoing issue.
      • Academic Computing Testbed
      • Collaborative Spaces
      • High Performance Computing
        alexp: this has been tried in various forms on Athena over the years but has not been successful and was discontinued due to lack of customers. A likely reason is that it worked in "mainframe-batch" mode. A possible way to resurrect it would be to offer "desktop interactive supercomputing" a la NVDIA CUDA parallel processing on graphics processors. Math and related departments are currently quite interested in this.
      • High Performance Graphics
      • Virtualized Computing Environments
        alexp: this has a lot of potential but will likely be manpower-intensive in design and construction. Also a lot of help desk support may be needed... we're still getting up to speed on VMWare and I've run into lots of ways of breaking it. One single aspect of this that could be highly beneficial to many students is to make Athena 10 virtual machines available to them. This would provide access to the many benefits of Athena without requiring them to install Linux on their computers (many are likely to not want to do that for various reasons).
    • Various Different User Focus
      • All Students
      • Administrators
      • Researchers
        alexp: software vendor licensing restrictions come into play in the user categories above and are difficult to enforce. Another issue is allowed use on personal machines vs. MIT-owned machines only. Most students and faculty strongly prefer to compute on their own machines (particularly laptops) rather than going to a computing cluster or office computer. We try to negotiate this right in current software licenses to the extent we can. Vendors often have piracy concerns- these can be alleviated by using floating license servers to control access (as done in our student Matlab distribution).
      • Access restrictions based on DLC ownership.
        alexp: implementing restrictions and keeping them up to date is complicated and resource-intensive. This is often onerous to users; again license servers come into play. On Athena we've gone to substantial effort to not have to restrict by user/machine location/departmental affiliation, but rather to be able to let a limited pool of licenses "float" across all machines/users on campus.
      • Special hardware purchased by system owner

    Key OS Distributions:

    1. Redhat / CentOS / Fedora
      alexp: almost all the software we've provided to-date has been primarily for this group and we have a lot of experience with it. Almost all commercial software vendors certify for RedHat first, to some extent SuSe and with Ubuntu playing catch up. Virtually no one certifies for LSB. Applications rated for Redhat almost always work with at most minor tweaks on Fedora, and we can get a lot of mileage in writing up documentation for how to get our core applications running on the other members of the family (as in the "Key Functionality for Fedora" pages for our key components I wrote for the Linux self-help wiki).
    2. Debian / Ubuntu
      alexp: we're starting to get a lot more Ubuntu familiarity with Athena 10. A lot of users are turning to it these days. In my recent commercial software vendor survey of support for Ubuntu, I only found about 30-40% or so that say they support Ubuntu; yet I was able to get essentially all Athena apps running on it with mostly moderate effort. Google is invaluable in figuring out how to make things work, and often much more useful than vendor support. As mentioned above for Fedora, we could get a lot of mileage out of some "how to get it working.." local documentation for our key components. Ubuntu has the largest (and growing) collection of free native packages which is a substantial advantage.
    3. "bsd" which could include MacOS and Solaris
      alexp: is there much demand for "native UNIX" Mac applications as opposed to the standard Mac/Cocoa ones? Systems like Fink make it easy to deal with packages on the UNIX level similar to Linux distro packages, but do we want to get into supporting this sort of thing?
    4. Gentoo alexp: it's not clear how many local users there are (or of
    5. SuSe, below), and Gentoo's source build system is cumbersome and very slow; given limited resources and a bad economy, we should assess how many distros we can actively support beyond RedHat*/Ubuntu/Mac (which I think have the majority of users).SuSe: SLES/SLED/OpenSuSE
      alexp: Novell's partnership with Microsoft is having some interesting (and controversial) consequences for Linux access to Microsoft technologies, notably in the mono/.NET area.