The Future of Linux

14 July 1998

Prepared Question #1

• Jeremy Allison (Samba): I think he was one of the panelists who made the comment that ``Linux is a commercial OS''; his answer amounted to choice #1:
  • Linux is very standards-compliant (e.g., Posix); a good approach is to develop first on Linux, then port to proprietary Unixen. For example, Samba has three separate pieces of code to deal with some aspect of file-system stuff; Linux supports all three interfaces, so they just choose the one that runs the fastest.
  • He did give a wishlist of improvements he'd like to see, though: 64-bit file-system support (``for those 20 GB Exchange databases''); access control list (ACL) support; asynchronous I/O support; NFS file-locking and improved performance [amen to that]; and a thread model like Solaris has.
  • He noted that Linux has current support for more platforms than any other OS: x86, SPARC, Alpha, PowerPC, 68k, etc.
  • Bugs are fixed the fastest, especially security fixes.
  • There's been a Linux ``iBCS'' module to support SCO Unix binaries for a couple of years; at this year's Usenix, SCO announced (and/or demoed) a module to run Linux apps.

• Larry Augustin (VA Research): His answer was ``yes'':
  • Linux is not (yet) as far along as Solaris in supporting 64-way symmetric multi-processing (SMP). [I thought that the SPARC-based Fujitsu AP1000+ on which David S. Miller reported success last year was a big SMP box, but as pointed out, it's a distributed-memory multi-computer similar to the Connection Machine CM5.]
  • In his slides of user ratings (Datapro survey, which was mentioned several times during the evening), Linux was not only the overall winner in a field of half a dozen OSes (Windows NT placed last), it also won in all but two categories--and only Digital Unix was rated higher in those two (availability and performance). The other categories included reliability, technical support, price, etc.

• Robert Hart (Red Hat):
  • Linux is a commercial operating system. It is sold and supported commercially (Red Hat, Caldera and others); it is used commercially; its only difference is that the source code is freely available.
  • Key ``commercial OS'' features like a journaling file system and database access that bypasses the file system layer are coming very soon.
  • Many of Red Hat's users (more with every release) have never installed an OS, which means Red Hat has to ``reverse engineer'' their hardware configuration.
  • Why is Linux not ubiquitous? It's still not suitable for everyone (he mentioned his ``75-year-old mum''), and although there are good office applications for Linux, there aren't any killer ones yet. (There were follow-ups to the suitability comment by some of the other panelists.)

• Sunil Saxena (Intel): He presented some slides that amounted to a ``yes'' response as well:
  • Strengths: Linux is becoming the OS of choice of ISPs; on 32-bit Intel systems, Linux has broader device-driver support than any other Unix (e.g., SCO, Solaris/x86, etc.); its Open Source model means that updates, patches and bug-fixes happen in ``Internet time.''
  • Weaknesses: SMP support and scalability is still evolving (although he noted that did a successful two-day port to the brand-new, four-way Pentium II Xeon system that Intel and VA Research showed off); good server management is missing (e.g., using a remote serial line or modem to update things, including the BIOS); drivers for high-end hardware tend to be lacking; and large-memory support (say, multi-gigabyte range) isn't there.
  • Making it better: he said (and repeated several times throughout the evening) that Intel really wants to help and do more to support Linux, and in particular they see the following as likely areas of collaboration:
    • more than 4-way SMP (serious scalability, at least 16 to 32 processors)
    • drivers for high-end platforms
    • direct server control and management
    • support for PII features such as 36-bit addressing (up to 64 GB of RAM), enhanced system calls and save/restore, MMX instructions, the page attribute table, and on-chip performance monitors

• Linus Torvalds: He started off with a comment to the effect, ``What can I say? I came to listen to the others.''
  • He noted that Linux originally was a one-person OS; it was never intended to be useful to others. He also pointed out that it has just shown up on the list of the world's most powerful supercomputers (in a cluster design); he thought it made #316. [actually #315 in the June 1998 list--see this press release for details] At the other end of the spectrum, it's being ported to Palm Pilots. ``I don't see that [its broad portability and usefulness] ending any time soon.''
  • He responded to a couple of Sunil's comments:
    • 36-bit addressing on Intel: ``We've been doing that on Alpha for a while.''
    • Page Attribute Table: he didn't know about Intel's implementation but said that he'd suggested it to an Intel engineer a few years ago; ``I don't know if they did it right, but if so, I'll be happy to use it.''

Copyright © 1998 Greg Roelofs.