lkml.org 
[lkml]   [1998]   [Feb]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: VFS 64-bit clean? Not yet (was Re: large-file-system)

>> The rest of us would like to NOT see performance go to hell.
>> It would be nice to be able to master a DVD disk. That requires
>> only 2 to 4 bits more than we have now. Going with signed sectors
>> would be 9 more bits. Remember, this is exponential expansion.
>
> [DVD uses UDF v1.2]
> You need 47-bit inode numbers for UDF: 16-bit partition,
> 31-bit block. If you throw in an offset too, that's another
> 11-bits (2k blocks) at least.

So tell me, how many UDF filesystems are larger than a terabyte?
Of those, how many are directly connected to 32-bit hardware?
It really doesn't matter what the theoretical filesystem limit is.

This is _not_ an external API. This is _not_ an on-disk format.
Compatibility is not an issue, so this can't lock us in like the
old 640 kB problem or the year 2000 problem. There is no need to
define something that will last until the end of time.

Intel will ship a 64-bit chip in 1999. 32-bit signed sectors (1 TB)
should be good until about 2010, and the Alpha could do 64-bit today
if the integer type is "long" instead of "__s32".

BTW, the UDF addresses remind me of C-H-S sector addressing.
It is brain damage. The addresses ought to be linear.


For those that don't quite get what "terabyte" implies,
here is info on EMC's Symmetrix 5700 with 96 disks:

data 1093 GB
cache 4 GB (non-volatile of course)
size 6x6x3 feet
weight 3106 lb (you must have a raised floor too)
power 6.19 kVA
heat 21102 Btu/hr

This unit is redundant, with hot-swap microcode & parts.
If something goes wrong, the unit calls the maintenance
people all by itself.

It keeps 16 Fibre Channel connections going simultaneously,
which is 1600 MB/s (12.8 Gb/s). You can do 32 UltraSCSI channels
instead if you prefer. It also supports ESCON, whatever that is.

Price? Oh, I think the 64-bit CPU will look like pocket change...

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu

\
 
 \ /
  Last update: 2005-03-22 13:41    [W:0.065 / U:0.048 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site