Re: NAND Based Flash RAID
- From: "Folkert Rienstra" <see_reply-to@xxxxxxxx>
- Date: Mon, 12 Dec 2005 22:52:20 +0100
"craigm" <none@xxxxxxxxxxxxxx> wrote in message news:Ggjnf.25$It6.1@xxxxxxxx
> Eric Gisin wrote:
> > "John Turco" <jtur@xxxxxxxxxxxxxx> wrote in message news:439D1F1C.2BE35E7B@xxxxxxxxxxxxxxxxx
> > > Eric Gisin wrote:
> > >
> > > > Absolutely clueless.
> > > > Any solid-state HD uses ECC and remapping, just like magnetic drives.
> > > > If it is flash-based, it also uses wear-leveling: http://www.lvr.com/mass_storage.htm
> > > > Flash drives have taken over mobile drives in extreme environments.
> > >
> > > Hello, Eric:
> > >
> > > Well, here's a quote from the Web page you just cited:
> > >
> > > "Wear Leveling
> > >
> > > Flash memory is a popular choice for smaller drives. Flash memory can
> > > withstand a limited number of erase/write cycles, however. A typical
> > > guaranteed number of cycles is 100,000. Firmware that repeatedly writes
> > > to the same addresses will quickly wear out that portion of the memory.
> > > To extend the life of a Flash-memory chip, firmware can implement wear
> > > leveling, which uses various techniques to access all areas of the memory
> > > equally."
> > >
> > > Nowhere does it imply that this "life extension" can allow a flash-based
> > > "disk" to replace a HDD, in a computer...which, for example, would
> > > access it far more often than a digital camera.
> > >
> >
> > My 4 year old SCSI drive has 3 billion writes. It has 30+ million sectors. Do the math.
> >
> >
> > > No problem, though, for a (considerably more expensive) RAM-based SSD.
> > >
>
> However, that would give you an average,
Exactly.
What you expect the number of times every cell to be
written to if wear leveling is doing what it claims to do.
> not the highest number of writes that any one block may have seen.
Doesn't matter.
> Some sectors were probably written once. Others will have been written much more.
> (Think about the directories and file system areas.)
That's not what he mentioned those numbers for.
>
> To do wear levelling, the controller need to keep a map of every block written.
And a lot more too. It also needs to know where that block is currently.
That's why they have "Dynamic mapping of any logical sector to any physical sector".
> How big do you think that is and where would you store it?
Are you questioning wear leveling?
> Where would you put this feature? In the host software or the 'drive'?
Drive obviously.
>
> Do you realize that FLASH needs to be erased before it is written?
What has that got to do with wear leveling.
> That also slows things down.
Not if you use up all the free blocks (pre-initialized, ie erased) first
(data unit rotation?) and erase the previously abandoned blocks in
times that the memory is not being addressed (garbage collection?).
>
> There are reasons why nobody does this.
So you also question John's quote
"Flash drives have taken over mobile drives in extreme environments" ?
> Ever wonder why?
No.
>
> craigm
.
- Follow-Ups:
- Re: NAND Based Flash RAID
- From: John Turco
- Re: NAND Based Flash RAID
- References:
- Re: NAND Based Flash RAID
- From: jbm666
- Re: NAND Based Flash RAID
- From: craigm
- Re: NAND Based Flash RAID
- From: John Turco
- Re: NAND Based Flash RAID
- From: Eric Gisin
- Re: NAND Based Flash RAID
- From: John Turco
- Re: NAND Based Flash RAID
- From: Eric Gisin
- Re: NAND Based Flash RAID
- From: craigm
- Re: NAND Based Flash RAID
- Prev by Date: I can't take that stupid flashlite!
- Next by Date: Re: HDD reads slower and slower...
- Previous by thread: Re: NAND Based Flash RAID
- Next by thread: Re: NAND Based Flash RAID
- Index(es):
Relevant Pages
|