xfs
[Top] [All Lists]

Re: Linux XFS write performance

To: delusion@xxxxxxxxxxxx, linux-xfs@xxxxxxxxxxx
Subject: Re: Linux XFS write performance
From: Michael Loftis <mloftis@xxxxxxxxx>
Date: Sun, 17 Apr 2005 13:31:44 -0600
In-reply-to: <2254.67.49.24.45.1113753740.spork@xxxxxxxxxxxxxxxxxxxx>
References: Your message of "Sun, 17 Apr 2005 00:10:57 EST." <2ca133d205041622106b61c016@xxxxxxxxxxxxxx> <31874.1113718752@xxxxxxxxxxxxxxx> <2254.67.49.24.45.1113753740.spork@xxxxxxxxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx


--On Sunday, April 17, 2005 9:02 AM -0700 delusion@xxxxxxxxxxxx wrote:


Hi,

Has anyone had experience/success getting 600MB/s or more (seq writes)
off XFS via a raid card like the Areca 16port sata?
The card does a 128k stripe and has 1GB of ram but I can't
seem to get the right xfs setup and write patterns.
Any tips getting the best sequenial write performance?

Which card specifically? Without PCI-X or PCI-Express (and you'd hav to be an x2 or x4 slot to approach this) (don't confuse the two, they're different) you're not going to achieve it. PCI has a hard limit around 133MB/sec. for 32b, 33mhz. 32b,66mhz gets you to around 533MB/sec. PCI-X is good to about 1033MB/sec, with 2.0 variants doubling that. PCI-Expres x1 gets you to around 250MB/sec. These are only theoretical maximums based on the signalling bandwidth. Reality eats away at those numbers considerably. Not to mention most CPUs would have a hard time reliably moving that much peripheral data around too.

You're in I/O territory only touched by dedicated systems, and by real servers, not PC hardware...although...if one were VERY careful about the design, it could be achieved (and we're talking figured in bytes, not bits here...if we're talking bits, then it's definitely doable)

Thanks!






--
Undocumented Features quote of the moment...
"It's not the one bullet with your name on it that you
have to worry about; it's the twenty thousand-odd rounds
labeled `occupant.'"
  --Murphy's Laws of Combat


<Prev in Thread] Current Thread [Next in Thread>