LISA05 Blog
From Computing Project Repository
| Table of contents |
LISA'05 Blog
On Friday, December 2nd 2005, four brave members of the staff at Oslo University College depart for San Diego. Their objective:
- To attend the Large Installation System Administration Conference (LISA '05) (http://www.usenix.org/events/lisa05/).
You can follow their exciting journey, as they eat, sleep and inhale the adventure that is LISA. Watch this space to get updated information as the events unfold.
Day 4, Tuesday
Below are the DRAFT notes from this year's workshop. Thanks to everyone who was able to make it. Please let me know of corrections.
Thanks, Steven
9-10 Introductions
- Problems
* How to connect our graduates to jobs
* Scott: How to build good problem-solving instead of recipes
* David: no money for course development -- drives towards
shared platform -> agree on a platform so we could share exercises
more easily
* Elizabeth: testing & assessment; external validation of testing
and assessment: SAGE certification standards/tests
* Matt: advocacy of education efforts
* Steven: how to connect us to others (in particular, to ACM)
* Steven: no web & ftp teaching materials
- Issue that seems common: lack of Windows administration expertise
10-12 Virtualization
- mln "My Linux Network"; Kyrre runs close to 100 hosts with mln. Built
on UML, so there are some stability issues because UML is still heavily
under development
- Two other products in the same space: vulab & vcml (by Zach Lowry: built on
QEMU); VMware or VirtualPC-based solutions are also possible.
- lab issues: logistics (physical space, setup time) x
cost (of hardware & space) = pedagogical payoff
- A solution adopted by everyone would need to be low/no cost and
cross-platform (e.g., VMware, VirtualPC, etc)
Assessment/Testing/Evaluation
- 'group riders' -
* weight group-work less than individual metrics (ie, homeworks 20%,
tests 80%)
* exams-per-assignment (ie, grade per topic is a combination)
* inter-group evaluations/peer evaluations
* add a notebook component per person
- Automated assignment grading
* some are doing it, basically by pre-setting test systems
* Acuitus handles tracking by looking at the following:
- machine's behavior,
- change ontology (ie, machine state),
- student belief state (ie, what the student thinks the problem is)
- David does interviews/demonstrations with each group of students to
do assessment
- David is moving _towards_ recipes _because_ of the 'just google it'
phenomenon (i.e., they search for 'email configuration' and find AIX
docs, even though they are working on, say, Solaris)
- Another common strategy is to give recipes at the beginning and
wean them from them over the course of the program.
- Some do pre-tests and post-tests to show improvement (this is
basically a required approach for accreditation). Some would like to
do 'test-out' approaches for earlier labs, for example. Using such
tests for diagnostic purposes seem to be a good theme. It might be
useful to have a common (ie, across this group) diagnostic test, or
perhaps a modular set.
- Big issues for us: (our) time, dealing with 'out of the ordinary'
approaches for solutions (e.g., instead of fixing the 'broken' box,
break the correct one
to match the broken one), breaking down larger projects into
(checkable) milestones,
- David has a vi tutorial he gives
- Modules people would like to see:
* LDAP (Aeleen Frisch did an LDAP module at Oslo this Fall)
* ActiveDirectory (an equivalent module to the LDAP one)
* Basic stuff: cd, editors, etc
* (Kyrre) a way to more easily and accurately describe overlap
between courses among schools
* Backups (and restores): Curt has some materials that may be
sharable/helpful. Aeleen did a backup module for Oslo also.
* Authentication & Authorization
- External assessment
* a more generic tool like SAGE cerification would be viewed as a very
positive thing
* one view is that a particular exam (within a course or program)
is hard to be exchanged with an external one
* 'nice to have', not necessary
* could be nice as an integrative exam (within a 'track', 'thread',
'concentration' or 'emphasis' program)
12:30-1:30 lunch
1:30-3 ACM IT Curriculum (and community issues)
* sysadm-education recommended curriculum models
* How to connect our graduates to jobs
- bringing students to LISA has mixed success
- internships are a great idea
- Oslo has had some success where students' final
projects (which include an industry customer) ends up
in job offers for those students (ie, job offers from the
customer).
- visibility and credibility (of the field) is an issue
* Advocacy of education efforts
* How to connect us to others (in particular, to ACM)
* Mark suggests a consortium idea -- grouping together to
- apply for grants/funding
- publish papers together
- (collaborate in general)
- Consortium of System Administration
with the overall view of raising visibility of System Administration
education and research
* Oslo:
MS program:
- has graduated one class (of 12)
- second class has ~15 students
- next has ~15 students
- two goals of program: terminal degree _and_ leave open
option of research (ie, PhD)
- currently 2 PhD students, expecting to graduate in 2007
* Alva currently has 4 PhD students, Yizhan Sun (almost finished),
Ning Wu (post-qualifiers), Mark Chiarini (pre-quals, post-courses),
and Hengke Susanto (working on autonomics).
3-3:30 Break
3:30-5 Textbooks & Materials
* Quite a few _are_ using Mark's Principles book as the core
text (both w/in the workshop and outside)
* Scott's school has access to 'Books 24/7'
* Steven's school has access to Safari (safari.oreilly.com)
* Technical book vs textbook: Kyrre sees a difference in quality
of reports in the original sources (recipes vs textbook)
* Curt is using outside reviewers of student papers (eg, students
in a different course will give feedback on the writing)
* Firewalls and Intrusion Detection (Kyrre is using)
- Northcutt's Network Intrusion Detection
- Ziegler's Linux Firewalls
* Ethics: some have a few chapters, but there don't seem to
be any good Ethics texts for System Administration (The Gift
of Fire seems to be fairly popular). Curt shilled 'Ethics and
Computing' by KW Boyer. The SAGE Code of Ethics is too
general and high-level to be the basis for a course in Ethics
for System Administrators.
* Security: Bishop, Secrets and Lies
* Networking:
- Peterson & Davies
- Kurose & Ross - for undergraduates
- Fourazan - good on details
- Tanenbaum
- Leon Garcia's: Communication Networks
- Alva has a lecture called 'Breaking IP'
* Another missing area: how to do write about research.
- problem is missing the scientific context
- Steven needs to dig up the reference on research methods
for CS paper that he has lying in his office somewhere
* Unix Fundamentals:
- Glass' Unix for Programmers and Users
- Mark Sobel's book Unix System V
- It's hard to beat Kernighan and Pike
- David would be interested in collaborating on a book for this
- Mark offerred to let people start from his notes to put
together an online text/version
- Mark offered to let us host our notes & materials on the Oslo
'System Administration Resources' site so that we can more easily collaborate/share materials (http://www.iu.hio.no/SystemAdmin/). Contact Matt if we want to put content up.
- Another possibility is the Drupal site set up last year (or conference
wiki?).
- A site for sharing this information might be something useful for the
Consortium to have as a main(?) focus.
- Norway's Network of Excellence called 'Magix' -- they are motivated
to produce teaching resources. They are going to make a virtual laboratory to share the resources of ~30 universities. They might be able to assemble a scheme wherein the Consortium could access the resources as well.
* If someone is interested in looking at other student projects (e.g., Master's Theses), some way of sharing those might be something the Consortium could do.
- Another idea is cross-institution collaboration on projects.
- YAI is doing external audits of curriculum
Thoughts on this year:
* good discussion today * a paper idea would be to write up issues & challenges (and even a survey) of virtualization and System Administration, Security, & Distributed Computing Education (Kyrre and Matt, among others, are interested in that) * The consortium idea has some support. Oslo will set up a wiki. One goal would be to have it visually appealing (hint: anyone have a student project to do web design?)
Attendees:
Curt Freeland - curt@cse.nd.edu
Steven Jenkins - jenkinss@etsu.edu
Scott Orr - sorr@cs.iupui.edu
Elizabeth Zwicky - zwicky@acuitus.com
Steve VanDevender - stevev@uoregon.edu, Univ of Oregon
David Byers - davby@ida.liu.se
Kyrre Begnum - Oslo kyrre@iu.hio.no
Matt Disney - Oslo matthew.disney@iu.hio.no
Terry Countryman - GIT terry@cc.gatech.edu
Mark Burgess - mark@iu.hio.no
Alva Couch - couch@cs.tufts.edu
Rik Farrow - rik@usenix.org
Day 3, Monday
Notes from the configuration management workshop
at
http://homepages.inf.ed.ac.uk/group/lssconf/config2005/index.html
Day 2, Sunday
This was a day working on private things.
Day 1, Saturday
There really is not much happing today, as far as LISA is conserned. We just registered, and use the oportunity to get online as well. We used some of the time earlier today to do some work.
Simen's notes
Visualization can be used to improve understanding of complex matters. The idea is to create a simple interface that can be used to set up a simple scene. The framework can be controlled with only a few parameters. The interface can presented both as a 3-dimensional and a classic 2D graphical user interface (GUI). After the initial setup of the display, the visualization is controlled by modifying only a small number of parameters. The system will consist of two main parts.
- The visualization engine
- The plug-ins (monitors)
The visualization engine is just a simple interface with which the plug-ins will communicate. The communication will be done through a network socket. This allows for distributed plug-ins. It also allows for the plug-ins to be written in any language, as long as it can send and receive messages through the network.
Each plug-in will typically represent one input to the system. It may represent a single machine or network, or it may represent a property of a machine or network. By using several monitors, a complex picture of the whole system can be monitored with relative ease.
|
[edit] Mark's summary of our discussion
Visualization can cover everything from the model of policy in terms of graphs and promises to data observed and measured as consequences of those choices. Therefore we should combine models from reliability theory (fault/cause trees) with time series models and anomaly detectors, logging engines and controllers. Tie this to a cfengine user base. Combine a policy-based view of visualization with a control interface for setting policy? A good idea is to build the research around a prototype that will run on Unix and Windows. Then people (esp cfenigne users) will start to use this and we will get a whole bunch of feedback from real users. Policy-based visualization means being able to decide how events and states are visualized, so as to maximize the comprehensibilty of the data. Here are some ideas for visualizing complex data.
In terms of abstractions, we want to attach multimedia representations to:
[edit] Siri + Kyrre overlap projectThe only example of autonomous agents that is currently in widespread use is BGP. Make a promise and Maude simulation of BGP. How do promises get amde in Maude? Game payoff - how does it fit in? Payoff is a subjective function of the measurements. There is an analogy with routing - Metrics (or utility factor for decision-making) - Policy (bgp) OSPF etc are like legacy systems that simply assume cooperation tacitly. As a result they do not scale well etc. Compare how OSPF and RIP networks look in terms of promises. RIP only promises its nearest neighbours. OSPF promises to listen to any flooded message.
Kyrre and Siri should formulate a version of promises in Maude language so we can explain our work to the Oslo group in their own language. Promises can be viewed as logic, but the are also a stepping stone to methods of approximation. e.g. a promise may be subject to noise and fluctuations from the environment, thus, while the ideas might be logically consistent in spirit, the details could fall short of this idea.
|
Specifying networks and systems in Maude
Maude (http://maude.cs.uiuc.edu/) is a formal specification and analysis tool based on rewriting logic.
Pictures
The whole gang (except me, who is taking the picture) was joined by Æleen Frish for a dicussion about nothing in particular. Later we went on a litter excursion to somewhere and looked at some seals. It was nice weather. We also ended up eating lunch.
