for reasonably moderate size couple of datawarehouse ( 1.2 TB each ) servers with approx 70% read 64KB block size and 30%write 32 KB block size on RAID 0+1 disks, IOPs plays a significant impact : Bus bandwidth plays relatively small impact between fibre and SCSI since both can easily handle 200 MB/s per channel.
since #disks in SCSI is huge : 160 and since I/O still take 40% of query time, increasing spindles seems a logical idea, but then u have to add more SCSI controllers on non-competing PCI-X 64 bit 100/133Mhz slots which are limited within the servers ( usually 6-8 in quad servers ), will fibre disks give me the same #IOPs ( like 7.5-8K per database server) with halving #disks from 160 to 80 ( since fibre disks and controllers and switches costs with 2x current SCSI setup)????
lower#disks per database server on SAN gives me added opportunity to scale to higher #disks at a later date.
since all disks currently in SCSI are 73 GB 10K, will I keep the same performance for 80 73 GB 10K fibre disks from current 160 and will I be able to accomodate a fast SAN with growth upto 210 disks for under $150K with switches and HBA's ( I am assuming six-eight HBA's per database server)
but only if I get the same performance for approx half the #disks.
since #disks in SCSI is huge : 160 and since I/O still take 40% of query time, increasing spindles seems a logical idea, but then u have to add more SCSI controllers on non-competing PCI-X 64 bit 100/133Mhz slots which are limited within the servers ( usually 6-8 in quad servers ), will fibre disks give me the same #IOPs ( like 7.5-8K per database server) with halving #disks from 160 to 80 ( since fibre disks and controllers and switches costs with 2x current SCSI setup)????
lower#disks per database server on SAN gives me added opportunity to scale to higher #disks at a later date.
since all disks currently in SCSI are 73 GB 10K, will I keep the same performance for 80 73 GB 10K fibre disks from current 160 and will I be able to accomodate a fast SAN with growth upto 210 disks for under $150K with switches and HBA's ( I am assuming six-eight HBA's per database server)
but only if I get the same performance for approx half the #disks.
