DragonFly users List (threaded) for 2008-07
DragonFly BSD
DragonFly users List (threaded) for 2008-07
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: Hammer - block size

From: "Jim Wilcoxson" <jim@xxxxxxxxxxx>
Date: Sat, 12 Jul 2008 12:14:48 -0400 (EDT)

When I write 10-byte files, does Hammer use a full 64K block for each
file, and then if I extend/truncate a file but it stays < 64K, it stays in
the same 64K block on the disk?

Or does it pack multiple files into the 64K block, with file A being the
first 10 bytes, file B being the next 10?  Then when I extend file A to 20
bytes (and do an fsync), I get 10 unused bytes where A originally was
located, file B for 10 bytes, the new file A for 20 bytes, and the
reblocker comes along later and fixes this block when the amount of
fragmented space in the 64K block becomes excessive or the block is full
(but has holes)?


On Sat, July 12, 2008 3:23 am, Matthew Dillon wrote:
> :I've been reading the DF mail archives the last few days about Hammer,
> and
> :it looks very good!
> :
> :Today I saw a review of 1TB hard drives, and there are some interesting
> :charts on sequential I/O performance showing that the "sweet spot" is
> :128K, across many vendors' drives:
> :
> :http://www.xbitlabs.com/articles/storage/display/1tb-hdd-roundup_10.html
> :
> :On some drives, there is not much difference: 110MB/s for 64K vs 120MB/s
> :for 128K.  But with other drives (WD), the difference is 105MB/s for 64K
> :vs 185MB/s for 128K.
> :..
> :Good luck with your 2.0 release!
> :Jim
>      Damn.  Look at those prices!  So nice!  We're going to be drowning
>      in multi-terrabyte drives by mid-next-year, for sure.
>      I wouldn't worry too much about the numbers, the sequential
>      performance is almost irrelevant.  It's the random read and write
>      performance that matters the most.
> 					-Matt
> 					Matthew Dillon
> 					<dillon@backplane.com>

[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]