Subject: Re: twe status queries?
To: None <email@example.com>
From: Thor Lancelot Simon <firstname.lastname@example.org>
Date: 12/02/2005 20:45:15
On Fri, Dec 02, 2005 at 03:21:40PM -0800, email@example.com wrote:
> In message <20051202202439.GA11241@panix.com>,
> Thor Lancelot Simon writes:
> >The 7500/8500 series 3ware cards are pretty cheezy, IMO. The "R5 Fusion"
> >is just a stripe cache, and indeed it's only large enough to cache one
> >stripe at a time -- so they just give it to the first array you configure.
> >Their error reporting is not very good and their performance is just so-so.
> Isn't this is the family that was really hot about 5 years ago? The
> one where the Linux community raved about throughput of 100Mbyte/sec
> writes and 300 Mbyte/sec reads? (i dont recall NetBSD numbers, sorry).
It's not very hard to do 100MB/sec of writes to a stripe set without
parity! The one thing you can say for the 3ware cards is that they
kept them simple enough that it was feasible to take advantage of
high internal bandwidth, unlike more complex PCI RAID cards that do
much more, and pay with expensive data copies across slow/narrow
internal buses for doing so.
> What I'm really looking for is high-density PCI-e SATA, which AFAIbC
> oils down to Areca or Highpoint, which both have FreeBSD drivers, but
> not NetBSD drivers. (I think both are FreeBSD CAM drivers.)
I think you can expect a PCI-e card from LSI sooner or later. The PCI-X
card will work in a 64/66 PCI slot, if you happen to have one of those
It's frustrating that we don't have drivers for many of the new storage
cards that look "like" -- but aren't quite -- SCSI host adapters. One
reason for this is that in the past, we shared a lot of drivers with
FreeBSD, but with their switch to CAM it's a lot harder to do that (there
were, really, only two or three new SCSI host adapter designs in the
many years after the switch). Another reason is that these things are
poorly documented and often the FreeBSD or Linux drivers are the only
place many details about them are publically divulged at all. The
Marvell SATA/SAS combo chip (present on many server motherboards) is
a case of this: the Linux driver is a real gonkulator that makes both
SATA and SAS targets look like SCSI and it's hard to do much with it
even now that Marvell have finally released the source. :-/
Thor Lancelot Simon firstname.lastname@example.org
"The inconsistency is startling, though admittedly, if consistency is to be
abandoned or transcended, there is no problem." - Noam Chomsky