another newbie question

[Tony Hammitt]

Well, you have to remember that this whole thread started when a home
user asked about using an unsupported RAID card from Linux.  I would
assume that you would get better results from using a supported
kernel extension than from using such a card.

You have to put up with the fact that your Ultra160 card has to hook
up to the computer through one bus.  PCI32 (at 33MHz) is limited to
132 MB/s roughly divided by the number of cards on that bus.  Even if
you use PCI64 at 66MHz, it's still slower than the bandwidth that the
processors are capable of doing.  128MB of RAM is considerably less
than the 2GB of main memory.  1.7GHz of PIII is faster than the
processor on the card.  The operating system doesn't have to have the
data transferred into it before it can do the checksum calculations,
and it can use more than one bus to transfer the data to the disks.

I'm not saying that all hardware raid implementations are slower than
all "software" implementations.  I'm saying that by limiting yourself
to one processor, a smallish amount of memory and ONE bus connection,
you have limited the maximum performance of the system.  That limit
can be pretty damned high, but it's still a limit that's lower than
the theoretical limit of the entire system.

If your experience shows you that one of those cards is really fast,
just think how fast four would be.  You can get boxes with 8 Xeon's
and four PCI buses, then hook up four sets of these cards.  The only
way to get full striping across all drives (maximum performance) on
Linux is to use "software" raid.  "Software" in quotes because it's
really using the system's hardware at a very low level, with
full-speed access to all of the system resources.

But we were talking about how a _home_ user can use raid on IDE and
kernel v2.2.  If the card is not supported, then they have to use
"software". We still live in the world where driver support for Linux
lags hardware availability.  "Software" raid is probably going to
work better than someone's reasonably well educated guesses about how
the hardware works.

It's pretty likely that there is no reasonable facsimile of Linux's
"software" raid for windows.  So on windows, expect that the
non-disclosed card driver is going to beat a real software
implementation.  But I don't care about windows; that's some other
list.

Take care,

	Tony

Brian Kelsay wrote:
> 
> If you think this is true about the hardware RAID then you haven't seen the
> Adaptec 2100 Ultra 160 RAID with dual channel and up to 128MB of standard
> SDRAM.  We are testing a couple here on a server that performs backup and an
> SQL server.  Fast as snot.  It's great when money is no object on a project,
> "Just make it fast."  Those servers got Dual PIII-850 on one and Dual 600 on
> the other in a 2U rack case, server class board, nice AGP video and Seagate
> Cheetah 18 and 36GB drives.  2GB of RAM on each.  Gods, that's a lot.  The
> next generation of SQL server is going to get 4GB most likely.  I know
> Sprint probably has stuff to blow this away, but for us it's fast.
> 
*snip*

-- 
Doubt is not a pleasant condition, but certainty is absurd. -- Voltaire