AXISHK
asked on
Number of Disks calculation
See the scenario to estimate disk requirement. Any idea how to estimate the disk required, ie 28 disks in case 1 & 250 disks in case 2 ?
Scenario 1
A data warehouse that is used by a small number of users. Each query involves one hundred 64 KB random reads and little log activity. To support 25 users who all want one-second response time, the system would have to support 2500 I/Os per second (25 users × 100) or 160 MB/sec (2500 I/O per second × 64 KB).
A reasonable server configuration would have a dual-channel SCSI RAID controller on a PCI-X bus attached to a total of 28 disks using RAID5. Here, RAID5 is used because there is little write activity.
Scenario 2
Heavyweight OLTP application where each transaction involves 200 – 8 KB random I/Os and a 400 KB of sequential log writes. To support 100 transactions per second, the data disks would have to support 20,000 random I/Os per second (100 × 200) or 160 MB/sec, and the log would have to sustain 40 MB/sec. The total bus would need to sustain 200 MB/sec (160 MB/sec + 40 MB/sec). A reasonable server configuration would have the server connected to a SAN via three fiber channel adapters on at least two PCI-X buses. The SAN would contain 250 disks on RAID1 for the data files and four disks using RAID1 for the log files.
Scenario 1
A data warehouse that is used by a small number of users. Each query involves one hundred 64 KB random reads and little log activity. To support 25 users who all want one-second response time, the system would have to support 2500 I/Os per second (25 users × 100) or 160 MB/sec (2500 I/O per second × 64 KB).
A reasonable server configuration would have a dual-channel SCSI RAID controller on a PCI-X bus attached to a total of 28 disks using RAID5. Here, RAID5 is used because there is little write activity.
Scenario 2
Heavyweight OLTP application where each transaction involves 200 – 8 KB random I/Os and a 400 KB of sequential log writes. To support 100 transactions per second, the data disks would have to support 20,000 random I/Os per second (100 × 200) or 160 MB/sec, and the log would have to sustain 40 MB/sec. The total bus would need to sustain 200 MB/sec (160 MB/sec + 40 MB/sec). A reasonable server configuration would have the server connected to a SAN via three fiber channel adapters on at least two PCI-X buses. The SAN would contain 250 disks on RAID1 for the data files and four disks using RAID1 for the log files.
ASKER CERTIFIED SOLUTION
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
ASKER
Tks
ASKER
here are some some window counters measured on my SQL Server
Max IOPS at peak hour
Physical Disk 0 \ Disk Reads/s 117
Physical Disk 1 \ Disk Reads/s 2222
Physical Disk 0 \ Disk Write/s 102
Physical Disk 1 \ Disk Write/s 200
Throughput
Physical Disk 0 \Disk Read Bytes/s 2,746,925 (~ 2.7 MB)
Physical Disk 1 \Disk Read Bytes/s 170,586,869 (~ 171 MB)
Physical Disk 0 \Disk Write Bytes/s 1,659,712 (~ 1 .7 MB)
Physical Disk 1 \Disk Write Bytes/s 10,324,181 (~ 10 MB)
My concern is the throughput. How to evaluate the disks in order to cope with the throughput in peak hours ? How to estimate the RAID controller or bus to cope this this load ?
Tks