defending FOSS was Re: [Linux-aus] [Fwd: Re: [xxxxxxx] FW: Re: S.A. Parliament - Ubuntu Matter of Interest]
tconnors at astro.swin.edu.au
Mon Dec 4 18:37:01 UTC 2006
On Mon, 4 Dec 2006, Brenda Aynsley wrote:
> * ... it is currently used mostly for running non-critical or highly expensive
> infrastructure ... delivering non critical web content, internet services and
> other non critical internet-related services
> [maddog in part responded to this but other examples particularly in australia
> would be great]
My experience is that all Australian supercomputers are now primarily
(VPAC, APAC, Swinburne) or have large (ac3) Linux clusters on both exotic
and non exotic hardware.
They are quite critical, valued at about $2 per CPU hour, industry rates,
and most software is written such that a crash on one CPU of say, 32, will
result in the entire parallel job being taken down (damn you, MPI). In
non-checkpointed code, this could feasably cost a week * 32 CPU days in
compute time, or $10,000 (it is often not worth or possible checkpointing
code that might only run a comparatively short time). Having personally
burnt about 128000 CPU hours, I can vouch for the reliability of the Linux
clusters I have used (but not necessarily the mains power they were
connected to :).
Similarly, the sheer volume of data can sometimes make it impractacle to
do backups, with departments often living with the risk of losing months
of calculations or instrumental data (the Square Kilometer Array
radiotelescope, to have a 50% chance of being hosted in Australia, and its
precursor projects already hosted in Australia, come to mind). They need
a reliable filesystem, running under a reliable kernel. Linux and the
BSDs are demonstrably suitable.
> * Where more reliable services are required, it is more of Unix flavours which
> are more often used than Linux
> [what is meant by this do you think?]
I think they are mistaken -- yes, some other Unices have more fault
tolerance and hot swapping built in, which of course are definitely not
running under the x86 architecture. But more often used than Linux when
reliability is required? Not these days.
> * Costs [of implementing FOSS} include change of hardware, retraining staff,
> additional expert maintenance staff, buying online support 24/7, design and
> implementation of addtional redundancy to have 100% backup in case of
Depends what you are changing from. If you are changing from a MS shop,
it is highly unlikely you will need new redundant hardware -- Linux will
make better use of what you currently have. Other changes of hardware
even when changing from other Unices? Linux works on how many platforms
> [are there total cost of ownership current docs around on this?]
> * ... What happens when someone sneezes and the system locks up, or new
> development is required to address a bug or new feature? The FOSS zealots
> don't mention the hidden cost of open source adoption.
I have yet to come across serious software that crashes at the smallest
disturbances. (Mozilla is not serious mission critical software.)
If closed software has a bug, unless you have a *seriously* big contract,
good luck in getting a fix in a timely fashion.
In open sourced software, unless the problem is commonly experienced, good
luck also in getting someone else to fix it. But if *you* pay someone to
fix it, you have very good luck indeed of having it fixed.
More information about the linux-aus