Laptop Linux at MIT A Work in Progress

1
Laptop Linux at MITA Work in Progress

• by William Cattey
• Massachusetts Institute of Technology
• wdc_at_mit.edu

2
A 3-Part Talk

• Part 1 A quick introduction to Athena
• Part 2 Laptops
• Part 3 Daunting Challenges

3
Part 1 A quick introduction to Athena

• Vision
• Services
• The Athena Value Proposition
• Tension between UNIX-land and PC/Mac-land
• Whither PC/Mac solutions
• Athena porting experience
• The Value of Linux

4
Vision

• Back in 1983, the vision was to scale the kind of
  high performance computing done on minicomputers
  so that it could eventually be deployed to
  hundreds of individually owned systems.
• Athena developed technologies and procedures to
  convert UNIX from a "one wizard per computer"
  model to a "one wizard per platform" model, and
  to deploy over 1000 systems identical enough to
  be maintained inexpensively, but different enough
  to be responsive to customer needs.

5
Services

• Kerberos authentication Athena brought you
  kerberos authentication that enabled secure
  communication on an insecure network populated
  with insecure machines.
• Zephyr client-server instant messaging Athena
  Zephyr does not have the pretty user interface of
  other systems, but was the first client-server
  instant messaging system and remains the leader
  in server scalability and clear security and
  authenticity of messages.
• Hesiod lightweight directory access protocol. We
  did LDAP before LDAP did.

6
Services cont'd

• Enterprise-wide file service We started with
  remote virtual disk, replaced it with NFS to get
  platform independence, and migrated to AFS to
  achieve greater security and scalability. We
  didn't invent enterprise file service, but we
  learned the real requirements and helped drive
  the market in the right direction.
• Install/Update Through 20 years of the school
  of hard knocks we evolved a software deployment
  system that could accommodate multiple hardware
  vendors, multiple operating systems, application
  software, and non-disruptive system updates.

7
The Athena Value Proposition

• The Benefit Hundreds of systems cloned from a
  master software archetype that automatically
  receive non-disruptive updates.
• The Cost Adherence to open standards, and
  awareness of issues of security and scale.
• The Challenge Making sure that customers do not
  feel constrained by the parts of the system that
  "need to be the same for all".

8
Tension between UNIX-land and PC/Mac-land

• The choice between Microsoft Windows and Linux on
  the laptop is informed by the mix of UNIX,
  Microsoft Windows and Mac operating system use at
  MIT.

9
Whither PC/Mac Solutions

• PC's and Macs were originally ignored because
  they were expected never to become powerful
  enough to matter. Instead they evolved to a
  broader base of software, more graceful user
  interfaces, and larger installed base than ever
  expected by the Athena founders.
• UNIX was surpassed by PC's and Macs in customer
  mindshare, diversity of applications, and ease of
  use by novices.
• PC's and Macs evolved to match but not surpass
  UNIX in networking.
• PC's and Macs have not yet caught up to UNIX,
  particularly Athena in the realm of large scale,
  low cost, maintainable deployment. We believe
  this is because the scalability value proposition
  of Athena has been ignored by an industry focused
  on other goals, and by a clientele unaware of
  what is possible. PC's and Macs come from a
  history of individual ownership - UNIX from a
  history of central administration.

10
Athena Porting Experience

• Athena has most easily been ported to platforms
  adhering to open standards such as 4.3 BSD Unix,
  Linux, and Solaris. Platforms minimizing
  vendor-specific value added proved easiest.
• Difficulties arise when interfacing to
  proprietary versions of services provided by
  Athena.
• Microsoft Windows has been the most challenging.
• SGI IRIX was next most challenging with its
  proprietary install system, inst, and its
  volatile code base.
• IBM AIX was next after SGI with the proprietary
  smit, hpfs, and HFT interface.
• There is no longer an IRIX or AIX Athena.
• There is a Windows Athena, but it takes more
  work than UNIX Athena to keep running, and to
  integrate new OS revisions.

11
The Value of Linux

• Athena likes Linux. It's open, generic, and
  easiest to interface to. Compared to other
  platforms, its code base is actually reasonably
  conservative and stable where it matters to
  Athena services.

12
Part 2 Laptops

• The Laptop Value Proposition
• Athena Work
• Disconnected Operation for Athena
• Services Affected
• Architecture and Approach
• Solutions
• Open Issues

13
The Laptop Value Proposition

• The Benefit Mobility. Wireless nomadic
  computing was recognized as valuable from the day
  Alan Kay expressed the vision of the Dynabook.
• The Cost More individual responsibility for
  maintenance and administration. When this cost
  goes unpaid, users experience catastrophic system
  compromises, failures, and data losses.
• The Challenge To find appropriate
  implementations of Athena lessons learned to
  reduce maintenance and administrative burden with
  corresponding improvements in reliability.

14
Athena Work

• The Athena Release Team helped qualify
  configurations recommended to customers for
  purchase.
• Modifications to the generic Athena Linux install
  were made to accommodate Laptops.
• User documentation for what to buy, and how to
  get Athena working on it was published.
• The issues of sometimes disconnected operation
  were reviewed and began to be addressed.

15
Disconnected Operation for Athena

• Up to now, Athena assumed the net was always
  there.
• Could a way be crafted to continue to offer
  Athena services, particularly system update in
  the context of a sometimes disconnected machine?
  
• PC and Mac disconnected operation is thus far
  better than Athena disconnected operation.
• The Athena disconnected operation project
• http//web.mit.edu/disco-athena/www/
• surveyed the issues and began crafting solutions.

16
Services Affected

• 1. Name service
• 2. Email delivery
• 3. Network printing
• 4. Kerberized login
• 5. time synchronization
• 6. Auto update
• 7. Software delivery
• 8. Zephyr instant messaging
• 9. AFS File service
• 10. Backup

17
Architecture and Approach

• We crafted an architecture for detecting network
  up/down events and forwarding those events to
  system demon for root-level tasks, and to a
  per-user demon for user-level tasks.
• We do a little bit of fiddling with the
  implementations or configurations of certain
  services.

18
Solutions

• 1. Name service -- We hope to add a fast DNS
  timeout hack.
• 2. Email delivery -- We configure sendmail to
  queue, and flush the queue with the system demon.
• 3. Network printing -- Version 1 rejects network
  print jobs if no network present. Queueing and
  flushing are planned for version 2.
• 4. Kerberized login -- Use local accounts, and
  hook kerberos ticket renewal into the user demon.
• 5. time synchronization -- hooked into the system
  demon.
• 6. Auto update -- break into checkpoint-able
  chunks.
• 7. Software delivery -- local disk is your
  friend.
• 8. Zephyr instant messaging -- hook into the user
  and system demons.
• 9. AFS File service -- fast timeouts added.
• 10. Backup -- Remind users to connect up and do
  it by hand.

19
Open Issues

• Detection of network up and down events still
  kludgey.
• Linux lacks the graceful tools to
• switch network interfaces.
• signal applications of shift to working offline.
• offline operation of network filesystem.

20
Part 3 Daunting Challenges

• Difficulties with Laptop Linux
• Issues in Linux support
• Athena's Recommendations for Laptop Linux
• A Constantly Shifting Landscape
• Next Steps

21
Difficulties with Laptop Linux

• The biggest difficulty is in handing users
  something that just works out of the box.
• Just identifying systems for which a working
  release, and user documentation can be produced
  and kept up-to-date is a major challenge!

22
Issues in Linux support

• Laptop systems often lack Linux support for
• current wired network chipsets.
• current wireless network chipsets.
• current graphics chipsets.
• current sound chipsets.
• power management.

23
Athena's Recommendations for Laptop Linux

• In order to reap the benefits of the Athena value
  proposition on the laptop platform, the Linux
  support must be generic, not proprietary.
• The current Athena Linux laptop offering involves
  extra hand-tooling to enable coexistence with a
  Microsoft Windows partition.
• Continuing to offer an Athena Linux laptop
  product requires a sustained commitment to Linux
  support from the hardware vendor.

24
A Constantly Shifting Landscape

• The IBM T30 of 2002 was a nearly ideal platform.
• This year's product line seems to have gotten
  most things right, but losing the ability to do
  Linux wireless networking owing to no drivers for
  the Centrino wireless chip was disheartening.
• We constantly worry in about the working
  relationship between graphics chip manufacturers
  and the XFreee86 developer community.
• The PC market and particularly the laptop market
  looks likely to continue to have short life
  cycles for the underlying chipsets.
• Microsoft coordinates closely with chip
  manufacturers, but the vendors of hardware that
  might run Linux have a variety of relationships.

25
Shifting cont'd

• It is very difficult for a customer to have clear
  expectations in the face of changing hardware,
  and frequent regressions in functionality under
  Linux.
• Can more be done to improve on this situation by
  either identifying a particular hardware
  configuration for longer lifespan, or by bringing
  more resources to bear to create closer
  coordination between the Linux software suppliers
  and the chip suppliers?

26
Next Steps

• MIT continues to evaluate the cost/benefit ratio
  for
• UNIX vs Microsoft Windows
• Laptop vs Desktop
• The Athena Release Team has a lot of UNIX
  expertise. We do Linux today. Will the hardware
  support make that effort continue to be cost
  effective, or will we find an easier time with a
  different UNIX solution that embodies the IBM
  PowerPC microprocessor in a laptop...