On 04-Mar-98 Wilko Bulte wrote:
> Interestingly enough these benchmarks are mostly run with direct attached
> disks, no RAID controller in sight. Assuming enough CPU horsepower
> (a AlphaServer 8400 will do just fine with 8x 625Mc CPU will do...) you
> do striping on the machine. Lots faster than any RAIDbox.
A lot of that has to do with politics. Some years ago i worked (briefly!)
for a certiain CPU manufacturer who ran RDBMS menchmarks mainly on certain
O/S and certain RDBMS engine (both made by the same vendor, dare I say).
One brand of RAID controllers was not used because it had a politically
incorrect CPU on board. the other brand of RAID controllers was pushed
hard as it had a politiclly correct CPU, but was so buggy, we could never
finish formatting the disks. We ended up with another brand of non-RAID
controllers which was politically acceptable (they did not use the proper
CPU or any CPU at all, but the politically wrong O/S was not getting the
inside track to this controller).
Yes, a very hot CPU arrangement, with many channels, on a fast I/O bus will
deliver carefully configured benchmark very well. It is not necessarily an
optimal setup for most of us (utilizing 5% of a disk is not what we
normally do either. There ar ebenchmarks, and there is real life.
> Who cares about data integrity, it is TPC you want for the glossies.
Actually some benchmarks are supposed to consider that. But I could tell
you some funny stories there too.
Shimon@Simon-Shapiro.ORG Voice: 503.799.2313
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message