cvs commit: src/sys/conf Makefile.sun4v files.sun4v options.sun4v src/sys/sun4v/compile .cvsignore src/sys/sun4v/conf DEFAULTS GENERIC GENERIC.hints MAC Makefile NOTES src/sys/sun4v/include _bus.h

mjacob at freebsd.org mjacob at freebsd.org
Tue Oct 10 22:28:25 PDT 2006


> One of the promises of SAS was that you could combine the benefit of SAS
> infrastructure (multipathing, channel bonding, and out-of-the-box
> connectivity) with the cost effectiveness of SATA.  Of course, that
> already existed with Fibre Channel, so the migration to SAS hasn't been
> as swift as many had hoped.

Storage purchasers and vendors are incredibly (over)conservative. That 
plus the fact that not even the SAS vendors (to my knowledge, as of 
about 9 months ago- stale info by now) have demonstrated working setups 
that have more than 4 SAS expanders. That *also* plus that fact that SAS 
drives are unattractive from a $/GB perspective while SATA drives are 
unattractive from a random IOPS perspective. This all combines to make 
people not *quite* ready to run with a SAS infrastructure.

Before I left DataDomain I was looking into various interconnect 
technologies that get one out of the capacity of 12-16 drives with a CPU 
head. Up until now, this has been dominated by Fibre Channel as the only 
real choice, but dual path SAS connections through ~7 expanders to make 
a rack seemed to be a good bet, with SATA drives for apps that are 
driven by density, SAS drives for apps driven by IOPS. That gets you, 
for SATA drives, something on the order of 90 spindles, or something 
around 30-35 TB per rack what with sparing and RAID5 or RAID6.

I expect that this might become attractive for the smaller data center 
stuff or where you can fit an app that actually can *use* all that data 
in one CPU head. FibreChannel is still going to be the choice where you 
have to interconnect more than one CPU head with your storage, or you 
have larger (e.g. > PB) needs. *Hopefully* iSCSI, or a less hampered
replacement, will come in with 10Gb and kill Fibre Channel, but I don't 
expect that for a few years.

> I suppose that a PATA drive should work on a SAS controller if a SATA
> converter is used.  It's also possible to put a Chevy Nova drive train
> into an Audi, given enough effort with a welding torch.  I just wouldn't
> recommend it.
>

It's a way to be able to flash SATA drives with new f/w using a Laptop 
and IDE PCI-Card connector.


More information about the cvs-all mailing list