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

List:       reiserfs-devel
Subject:    (reiserfs) RE: (reiserfs) Re: nvram
From:       Chris Mason <clmsys () osfmail ! isc ! rit ! edu>
Date:       1999-08-12 15:27:20
[Download RAW message or body]

Ok, I was not thinking of power conservation.  With proper buffer cache
tuning, the journal could help some...

Anyway, I've been thinking about the right way to make the journal tunable
on the fly.  How about a /proc interface for changing things like the max
transaction size/age/num_writers, journal size/location/device etc.
Changing some params would just require a commit of the current transaction,
others would have to flush all journaled blocks to disk.

It will be a while before I get that far, but apmd should be able to use the
/proc interface to switch the journal back and forth.

-chris

> -----Original Message-----
> From: Russell Coker [mailto:russell@coker.com.au]
> Sent: Thursday, August 12, 1999 7:35 AM
> To: Ronald Pottol; reiserfs@devlinux.com
> Subject: (reiserfs) Re: nvram
>
>
> On Wed, 11 Aug 1999, Ronald Pottol wrote:
> >Now that I think about it, I know you can get IDE flash ram disks, for
> >primary storage they are insanely expensive, but for a 32MB disk for
> >logging they may well be reasonable.
> >
> >The PC Card stuff is reasonable, 8MB/s is not a thrilling
> transfer rate, but it
> >is fairly good, and seek time is hugely better.  For desktop users,
> >and IDE (or SCSI) device is better, as it is on a better bus, and they
> >do not have to buy a PC Card adapter.
>
> I think that the situation is that we have two different user-requirements
> which result in the same developer requirements for logging to a different
> device.  Users of laptops want writes to go to a low-power
> device.  Users of
> desktops want writes to go to a really fast device.
>
> >Agressive cacheing, and writing to a journal file on a PC Card with
> >the disk spun down, spining the disk up only for reads or if the
> >journal is full or has been around too long, ought to help notebook
> >battery life a good bit.  Also, flushing the journal on suspend or
> >hibernate would be good.
>
> Sounds great!  Also when apmd detects the mains power coming on
> it could tell
> the FS to put the low on the hard drive, and only use the PC-CARD for when
> there is no mains.  When we have mains power we don't need to save power!
>
> >Lastly, integrating this with the suspend to disk kernel patch might
> >be good.
>
> What integration would be required?  Surely if the kernel
> recovers after being
> suspended then it will have all it's mounted file systems etc and
> everything
> will be fine!
>

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