[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-ext4
Subject:    Re: [Ext2-devel] External journaling patch
From:       Daniel Phillips <phillips () bonn-fries ! net>
Date:       2001-07-31 21:39:26
[Download RAW message or body]

On Tuesday 31 July 2001 07:33, Andrew Clausen wrote:
> On Mon, Jul 30, 2001 at 04:51:55PM +0200, Daniel Phillips wrote:
> > I'd like to be
> > able to discover the true geometry of a drive, using statistical
> > methods to get a rough geometry by sampling the timing
> > characteristics in a few places, or an exact map, constructed by
> > laboriously analyzing the timing characterists of each track.
>
> Sounds hairy...  got any tricks in mind?

Well, for example, consider the following algorithm: turn off caching, 
pick any sector s, then start reading s, s+i from some large number of 
i, take an accurate timestamp on completion of each read, and plot the 
delta t's, just for the forward seeks.  If your disk has a single 
platter this will give a sawtooth pattern with an overall ramp up.  The 
distance between points of the sawtooth give you the track size(s), and 
the time between peaks gives you the seek rate.  (Sector skew will 
alter the results slightly, I haven't thought about it that far.)

Even if the disk has multiple platters, the sectors may well be 
distributed in such a way that it acts like a single platter.  If not, 
the sawtooth will be a little more complex, increasing by steps.

You don't have to map the whole disk this way - it would probably be 
quite sufficient to carry out the procedure at a few evenly spaced 
sample points.

> I really need to learn more about the protocols... but basic things
> like: do we know when writes complete (as opposed to transfered into
> some write-behind cache)  I hope we have this one, for
> transactions...

We really need to know such characteristics.  And once we know, we 
should make it plain to the vendors which we prefer.

> Anyway, is there anything good to read?  I could only find marketing
> propaganda (even on the IBM site, www.storage.ibm.com)

I think that basically sums it up.  Try googling for "head settle time" 
or something like that and you'll find some out-of-date material that 
is nonetheless interesting.

> > Instead of simple track-to-track and average seek numbers, I'd like
> > to measure the actual acceleration profiles for the head.
> >
> > If we have such a disk analysis tool we'd also expect it to tell us
> > whether features such as disable cache, write barrier, etc,
> > actually work.
>
> Yep.  I didn't know you could disable cache.  If we can't, are there
> any good ways to get the numbers?

If we can't disable the cache I'd consider the disk broken.  That said, 
it would still be possible to extract the geometry, just harder and 
slower.  Personally, I wouldn't bother, I'd just add it to a list of 
broken drives.

--
Daniel

_______________________________________________
Ext2-devel mailing list
Ext2-devel@lists.sourceforge.net
http://lists.sourceforge.net/lists/listinfo/ext2-devel

[prev in list] [next in list] [prev in thread] [next in thread]