On Thu, May 02, 2002 at 08:19:04PM +0200, Martin Dalecki wrote:
...
> 15 drives == 16 interfaces == 8 channels == 4 controllers
> with primary and secondary channel.
Usually using both master and slave on an IDE channel spells disaster
performance wise, and I would be surprised if the hotplug stuff worked
with this as well...
>
> He will have groups of about 4 drives on each channel wich
> serialize each other due to excessive IRQ line sharing and
> master slave issues.
Use 8 controllers for the 15 (16) drives.
>
> 8 x 130MBy/s >>>> PCI bus throughput... I would rather recommend
> a classical RAID controller card for this kind of
> setup.
Because RAID controllers do not use the PCI bus ??? ;)
The bus-overhead on RAID-5 is not too bad unless you specifically construct
a workload to make it so (writes-only, scattered so that the kernel cannot
cache stripes to avoid read-in for parity calculation).
Sure, the PCI bus will be a bottleneck, and PCI overhead alone will decrease
the real-world performance to somewhere below the theoretical PCI bandwidth
limitations, but don't let this blind you - 100 MB/sec sustained transfers
can still be "good enough" for many people.
By the way, has anyone tried such larger multi-controller setups, and tested
the bandwidth in configurations with multiple PCI busses on the board, versus a
single PCI bus ?
--
................................................................
: jakob@unthought.net : And I see the elder races, :
:.........................: putrid forms of man :
: Jakob Østergaard : See him rise and claim the earth, :
: OZ9ABN : his downfall is at hand. :
:.........................:............{Konkhra}...............:
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue May 07 2002 - 22:00:15 EST