ZFS obn FreeBSD hardware model for 48 or 96 sata3 paths...

Freddie Cash fjwcash at gmail.com
Tue Sep 20 18:45:13 UTC 2011


On Tue, Sep 20, 2011 at 9:01 AM, Bob Friesenhahn <
bfriesen at simple.dallas.tx.us> wrote:

> On Mon, 19 Sep 2011, Freddie Cash wrote:
>
>>
>> L2ARC has extreme bandwidth limitations as compared with RAM.  Be careful
>> what you wish for.
>>
>> For writes (7 MBps, I believe); there shouldn't be any limits on the
>> reads.
>>
>
> If (for example) an SSD is used with a 200MB/s read rate for the L2ARC,
> then the L2ARC is limited to 200MB/s (as compared with perhaps 10GB/s or
> 20GB/s for RAM).
>

Ah, yes, obviously it's limited by the hardware, but so is the pool.  :)  I
meant there's no artificial limits on reads from an L2ARC device, or writes
to a ZIL device.  In contrast to the write throttling for the L2ARC device.


> The L2ARC is really all about eliminating the access latency of
> rotating-rust but any device will provide far less bandwidth than system
> RAM.
>

Yes, L2ARC is definitely slower than RAM.  But properly selected/configured
L2ARC will be a heck of a lot faster/lower latency than the pool.  Hence the
ordering I gave originally:  ARC -> L2ARC -> pool


-- 
Freddie Cash
fjwcash at gmail.com


More information about the freebsd-fs mailing list