I posted the following a while back and we had a number of good discussions. The tests I had done on my client's machine had been done remotely and their computer center. I got to go to the facility and see the machine this past week and realized something I hadn't reported on previously. On the Mac Pro with Apple RAID card running RAID 5, the drives are eneterprise level SAS drives whereas the drives used on the Pegasus R6 were only inexpensive and slower SATA drives. This makes the 20% speed increase by the Pegasus even more impressive since it was using slower/cheaper drives.
I have a client with a reasonably large database (2/3's TB) that would like it to run faster. They are running on a Xeon Nehalem Mac Pro (MacPro4,1) with Apple RAID Card, RAID 5 over 4 drives. This is the Mac Pro made between March 2009 and July 2010 and it is running Snow Leopard 10.6.8 on FM! S 11.0v3, 16 GB RAM.
I just recently purchased a Mac Mini wi th Lion on it, 4 GB RAM, and a Promise Pegasus RAID with 4x1TB drives over a Thunderbolt connection and also RAID 5.
Technically, the Apple RAID reads/writes in the up to 250 MB/s range and the Pegasus reads/writes up in the range of 500-750 MB/s. So I was really hoping for a much faster database on this new Mac Mini system.
I have a script from the database using MANY relationships and calculations, none of which are imports.
I used exact copies of the database and ran them on servers using a schedule script, so no client was directly connected.
Mac Pro: 26 seconds, Mac Mini 21 seconds.
Mac Pro: 24 seconds, Mac Mini 22 seconds.
Mac Pro: 24 seconds, Mac Mini 21 seconds.
Mac Pro: 25 seconds, Mac Mini 22 seconds.
So the Mac Mini is always faster, sometimes up to 20% faster than the Mac Pro with RAID. This is an OK improvement and we're going to go with it, but I have to admit, I had hoped for a MUCH bigger improvement.
The big plus is that the Mac Mini Server with Pegasus was about $2000, making this a MUCH cheaper solution than the Mac Pro which had about $6000 worth of hardware.
We also tested this on client machines running FMP 11 over ethernet and were getting in the 20% improvement, but nothing more.
The next test I will want to do is compare this against SSDs which are much better at random data calls.