[GRLUG] NOT LINUX - hybrid drives

[Adam Tauno Williams]

Quoting Bob Kline <bob.kline at gmail.com>:
> On Mon, Nov 12, 2012 at 8:08 AM, Adam Tauno Williams <awilliam at whitemice.org
>> wrote:
>> On Sat, 2012-11-10 at 11:18 -0500, Bob Kline wrote:
>> > On Sat, Nov 10, 2012 at 11:03 AM, Adam Tauno Williams
>> > <awilliam at whitemice.org> wrote:
>> > Bob Kline <bob.kline at gmail.com> wrote:
>> http://www.amazon.com/gp/product/ref=pe_172730_26630760_pe_b2/?ASIN=B003NSBF32mainson spinning rust in the SSD  
>> case).
>> > >Conceptually, the "sticky bit" in file
>> > >permissions was intended for just that
>> The UNIX sticky bit is an entirely historic footnote.  In ~25 years I've
>> never worked on a system that implemented 'orthodox' stick-bit behavior
>> - or at least where using it was ever advised.  [mostly because it is
>> just a dumb idea]
>> If we want to grasp at vague equivalencies I'd consider LD_PRELOAD to
>> better resemble pre-load-the-good-stuff than the stick bit does.  But
>> neither of these are really equivalent to storage tiering since the
>> entire point of such tiering is that it is dynamic;  rather than the
>> sys-admin [or worse, the developer] trying to guess [and inevitably
>> being wrong] what data is upper-tier worthy.
> Wouldn't it be simpler to just say
> that things like the sticky bit are at
> odds with the system scheduler?

No, I don't think so.   It is simpler to say that trying to out-think  
the algorithms almost always fails.

> Which in turn are said to be
> notoriously hard to get right.

I disagree,  Modern schedulers are awesome;  it is extremely difficult  
to make them near-perfect.  They do a really good job, generally.

> Apparently one can even suggest
> today that a process be associated
> with a specific CPU in a multi core
> system, it won't necessarily happen,
> for the same reason.

Correct;  and this is usually done more to keep a process  
out-of-the-way (keeping other cores available) than it is to  
'accelerate' a given process.  Even then it is a questionable practice  
in almost all cases.  I've seen systems where syslog and the like were  
pinned to a high core number

> As for dumb idea, perhaps to the
> extent that Ritchie and Thompson
> were dumb, but saying that would
> take a fair bit of hubris.

Bull crap.  Nobody said they were dumb, don't equivocate.   Saying  
someone had a dumb idea and saying they are dumb is not the same  
statement and everyone knows that.

And I will not defer the High UNIX Clergy.  Vint Cerf has had 'dumb'  
ideas.  Linus has had 'dumb' ideas.  Alan Cox has had 'dumb' ideas.   
And I have no doubt they would all testify to their own dumb idea  
having-ness.  UNIX has some pretty twisted up constructs [/dev files +  
ioctl calls +  ...] which well demonstrate the humanity [non-divinity]  
of its creators [despite the nearly divine elegance that it's devotes  
like to attribute to it] (and I'm a UNIX fan).