NetBSD/doc/ROADMAP

#	$NetBSD: ROADMAP,v 1.2 2005/11/11 05:51:20 briggs Exp $

A high-level roadmap for NetBSD

This file contains a general map of where we would like to take
NetBSD over the next N years.  It is not highly detailed or overly
specific about each item.  There are several different "TODO" files
and "NetBSD Projects" lists in various places that contain some
more detailed plans.  This is the framework in which those projects
and plans are expected to fit.

As this is a volunteer project, there are no specific dates beside
these items.  These items may or may not get picked up in any order,
and the roadmap may change as technologies and perceived needs
change.

The roadmap, of course, is constructed in the context of the
Project's (broad) goals:

	* clean design		* stable		* fast 
	* clean licensing	* portable		* interoperable
	* conformant		* commercial-ready	* research-ready
	* hobby-ready

In general, we are headed for:

	* "State of the art" tools (current (and stable) GNU tools,
	  addition of Solaris's dtrace or similar functionality, kernel
	  core dumps on all platforms and post-mortem analysis tools,
	  performance analysis tools with support for hardware assists
	  like PMCs)

	* Support for all devices without encumbered code

	* Managed growth of the base system

	* Minimal GPL / LGPL code in the base system

	* Maximal performance without compromising portability

	* "State of the art" technology in the kernel and userland

	* No bugs, no security vulnerabilities

	* In combination with pkgsrc, a complete system for a variety
	  of users, administrators, and researchers: desktops, embedded
	  devices, servers, workstations, and portables

This is, by no means, a comprehensive list, and purposefully aggressive.
One of the many challenges will be to achieve excellence in each arena
we tackle and not settle for being a "jack of all trades, master of none."

The following, more specific, items are divided into rough categories:
	1. Platform independent kernel
	2. Platform independent userland
	3. Platform dependent kernel
	4. Platform dependent userland
	5. Other

If you'd like to take on a project, please record your name/email,
the date that you're claiming a project (or part of a project--if
a part, please specify the part), and an expected completion date.
This will hopefully avoid both duplication of effort and too many
or too-extended stalls.

1. Platform independent kernel
==============================
aa. Modular scheduler supporting CPU affinity and some soft RT features.
    The BSD scheduler is good for a wide variety of operating conditions,
    but some applications really want some soft real-time capabilities and
    multi-processor systems could benefit from taking processor affinity
    into account (including support for HT CPUs).

ab. Reduction of the giant lock
    There are several proposals for the best way forward on this, but
    we really need a couple of people with time to step forward and
    lead us here.

ac. Expansion of wedge support
    Complete the development of wedges and retire disklabels except
    where needed for compatibility.

ad. Volume management
    Allow us to grow, shrink, and move partitions (and, where possible,
    filesystems).

ae. High-performance, maybe log-based, journalled fs w/ snapshot support
    Addition of logs, journals, and snapshots to FFS is a lot, another
    filesystem could be cleaner and faster.

af. Expansion of ieee1394 support
    Where possible, fully support DV, disk, and network devices.

ag. Generic device hotplug support
    Support hotplug of all devices and busses that support it.  This
    should be divided into subcategories and does cross over some into
    platform-dependent areas.  SATA, SCSI, FC, USB, Firewire,
    PCI (PCI-X, and PCI-Express), etc.  There is some rudimentary
    support present, but it is far from comprehensive.

ah. Suspend and resume support
    We should be able to fully utilize suspend and resume on PCs, macppc,
    and anyone else who supports it in hardware (sparc, hpcsh, hpcarm, etc).

ai. Complete support for LWPs
    There are still vestiges of the kernel that predate LWPification
    and should be updated.  [ What other than ktrace? ]

aj. PTHREAD_CONCURRENCY > 1 support
    A single process that uses threads should be able to reliably
    utilize more than one CPU.

ak. AIO support
    POSIX aio_*() with full support for Asynchronous I/O (AIO) in the
    kernel.

al. Modern parallel port support
    Complete support for bidirectional and "advanced" functionality
    from parallel ports.

am. NFSv4
    Bring our NFS up to current standards.

an. Update the locking mechanisms in the kernel
    This requires some platform support.  A good bit of work is on the
    now-archaic "newlock" branch.  It requires some overhaul of
    cpu_switch/scheduler so that mutex_*(9) and ltsleep(9) can interlock.

ao. Review TCP/IP developments
    Look into RFC3168 (ECN, http://www.icir.org/floyd/ecn.html) and
    other "recent" and current TCP/IP research.  Adapt our stack to the
    more modern world.

ap. Kernel linker (ala FreeBSD's kld)

aq. CARP/VRRP
    Functionality is great, but there might be some concern here over
    Cisco patents.

ar. UDF filesystem support
    OpenBSD has recently added this.

as. RAIDFrame support for 3-way RAID 1

at. RAIDFrame support for RAID 6

au. NDIS
    Support for binary network drivers on x86 platforms.

av. More modern drivers
    We lack support for a number of more modern devices (PCI-Express,
    RAID cards, etc.) that are supported on other open source OSes.

aw. iSCSI initiator support
    We should be able to utilize iSCSI volumes.

ax. Run-time changable limits to SysV IPC
    Some of the limits for SysV IPC are hardcoded in the kernel
    configuration--these should be changable via sysctl.

2. Platform independent userland
================================
aa. Keep up with the X world
    If we are moving to an X.org base, we need to move forward with that
    and keep up to date.  This will require some platform-specific work,
    but is largely platform-independent, and involves making X.org cross-
    buildable.

ab. Reentrant libraries
    Make sure that all libraries are re-entrant and usable for threaded
    applications.

ac. gcc updates
    This requires some work to rework the gcc4 builds to work with BSD
    make(1) or update BSD make(1) or consider the unthinkable.

ad. gdb updates

ae. binutils updates
    Probably go along with gdb updates.

af. Better post-mortem debugging tools
    It would be useful to have something between ps/*stat/etc. and
    gdb with a core file.  Something, perhaps, like SysV(?) crash(8).

ag. Better 802.11 utilities and support
    To truly support mobile users, we need better support for scanning
    for base stations and affiliating with them.

ah. Internationalization
    Citrus, wide-char curses (SoC integration?), collation, localized
    printf with positional parameter support, time & currency
    support, etc.  NetBSD has a global user and developer base and
    our i18n support should reflect that.

ai. System packages
    In some fashion, we need to support system packages.  This is at
    least to allow for sane binary audits and binary patches.

aj. Provide support for binary packages in install
    We should have an integrated install that can install a desktop as
    functional as other free operating systems.  Without sacrificing
    the quick and clean basic installs that we have now.  An extension
    of sysinst might fit the bill.  Or a tool that's both invoked by
    sysinst and available on a running system.

ak. Native signing/privacy software
    This could be BPG (from SoC) or openpgp-sdk.

al. Userland version identification
    What binaries are installed?  Who really knows?  This relates at
    least somewhat to (ai).

3. Platform dependent kernel
============================
aa. Move evb ports to evb* if they're not already there (sandpoint)
    The existing evb* ports are kind of catch-all ports for eval boards.
    Some of the existing non-evb* ports really belong in the evb* category.

ab. m68k ports need to share more code
    Some progress has been made here in recent years, but there is more
    work to be done.

ac. Kernel core dump support for all platforms
    Some platforms (PowerPC ports, ARM ports, etc.) don't have full
    support for kernel core dumps and post-mortem debugging through
    libkvm.  This should be updated.

4. Platform dependent userland
==============================
ab. m68k should be able to share sets
    Some progress has been made here in recent years, but there is more
    work to be done.

5. Other
========
aa. More and better regression tests
    The suite of tests that we have now is limited.  We should expand
    these and provide systems (or manage a volunteer pool?) to run the
    tests on -current and release branches on a variety of platforms.
    ( Perhaps virtualize some with qemu or similar? )

ab. Native Java on multiple platforms
    Getting i386, amd64, sparc64, and PowerPC would be a good start,
    although PowerPC will require more work.  We have the go-ahead,
    but we need the people to work on it.

ac. Power control framework
    Related to suspend/resume support, we should have some framework
    for dynamically stepping processor speed, controlling fans, shutting
    down and/or powering subsystems to conserve power and keep the system
    with thermal limits.