How do you "know" how fast of a storage subsystem you need for a server?
A couple years back I bought a file server, Dell 2950 (2u) with a RAID 1 set for the OS and a three drive RAID 5 set for the data store and a global hot spare. They are Western Digital YS drives "enterprise SATA". Over time I slowly added misc roles to the server such as a 10 user access database, Windows Update services, and some online backups. Under normal operation it has about 3% CPU usage and 25% memory usage. Disk queue length is under 1 to a max of 4.
Turns out I under estimated the amount of IO needed when recovering from certain file corruption issues. These issues tend to creep up once a year and the server just dies on me during this time causing more corruption and the need for even more resources. CPU and memory resources are still not being taxed during this time.
The Average Disk queue goes into the 12 range and spikes up to 24 ALLOT during this time. Disk seconds/write goes up to .114 and averages in the .080 range. Things generally fall apart and I'm forced to move much data off the server in order to finish repairs. Downtime, lots of man hours, and chaos. In the end I move everything back and nothing changes.
So I know how my current setup performs and my internal drive count in the machine is maxed out. I'm confident that ditching the SATA drives for 15k RPM SAS will solve the issue for an acceptable expense. I'd like to have numbers to look at to support it. How can I know for a fact that changing drives will get me where I need to be? Maybe it's a case where DAS is needed if I can't fit the necessary spindle count internally.
If a server is going to have many roles and I know the amount of users that are going to be using it how can I know what kind of drive performance I'm going to need? I suppose this would be more pressing if this was a VMware machine running multiple logical servers.
I've taken the ability to add additional internal drives for granted until now.
Moved to appropriate forum - Moderator Rubycon
A couple years back I bought a file server, Dell 2950 (2u) with a RAID 1 set for the OS and a three drive RAID 5 set for the data store and a global hot spare. They are Western Digital YS drives "enterprise SATA". Over time I slowly added misc roles to the server such as a 10 user access database, Windows Update services, and some online backups. Under normal operation it has about 3% CPU usage and 25% memory usage. Disk queue length is under 1 to a max of 4.
Turns out I under estimated the amount of IO needed when recovering from certain file corruption issues. These issues tend to creep up once a year and the server just dies on me during this time causing more corruption and the need for even more resources. CPU and memory resources are still not being taxed during this time.
The Average Disk queue goes into the 12 range and spikes up to 24 ALLOT during this time. Disk seconds/write goes up to .114 and averages in the .080 range. Things generally fall apart and I'm forced to move much data off the server in order to finish repairs. Downtime, lots of man hours, and chaos. In the end I move everything back and nothing changes.
So I know how my current setup performs and my internal drive count in the machine is maxed out. I'm confident that ditching the SATA drives for 15k RPM SAS will solve the issue for an acceptable expense. I'd like to have numbers to look at to support it. How can I know for a fact that changing drives will get me where I need to be? Maybe it's a case where DAS is needed if I can't fit the necessary spindle count internally.
If a server is going to have many roles and I know the amount of users that are going to be using it how can I know what kind of drive performance I'm going to need? I suppose this would be more pressing if this was a VMware machine running multiple logical servers.
I've taken the ability to add additional internal drives for granted until now.
Moved to appropriate forum - Moderator Rubycon
