Suggestions on pool upgrade drives.

[jamiejunk]

We’re been using FreeNAS for a number of years and it’s been great. Some of the drives are getting pretty old and I’m wanting to cycle them out as part of a replacement plan.
Right now we have lots of 2TB and 4TB drives. We’re facing some budget cuts, so I’m trying to get get the drives that will work for us performance wise, but at the same time maybe cut some costs. Our storage needs haven’t gone up very much and I don’t really expect them to in the future.

It seems that maybe 6 or 8TB would work for us. The price per gig is cheaper than us doing 2 or 4 gig drives. We usually do mirrors, so getting larger but less spindles will cut down our performance.

My thought was to look at the disk operations graph history and see the max IOPS over time for drives in the different pools.
From there figure out what our max iops for each pool was and make sure we get at least enough spindles to cover that.

Is my thinking correct on this?

Also, in looking over the graph I’m noticing some weirdness. For example this graph for da3, which is a 7200RPM sas drive. In this graph it says it hit almost 500 IOPS, but I thought the max iops for that speed of disk was between 75 to 100. Does anyone know what that’s about?

I know that our ARC and L2ARC will have a big impact on real world numbers. But like I said, things have been going good. I need to cut down the budget, just don’t want to cut to many corners and have a problem.

 

[joeschmuck]

My thought was to look at the disk operations graph history and see the max IOPS over time for drives in the different pools.
From there figure out what our max iops for each pool was and make sure we get at least enough spindles to cover that.

Is my thinking correct on this?

It's not a bad way of thinking about it but IOPS is not throughput which is another factor.

I think you would need to provide your system specs/pool configuration in order to get a good answer.

What you can do is provide your system configuration to include the drive models and you also could perform a basic throughput/IOPS test on your current system if you like, compare that to your max IOPS on your GUI graph. This may give you a feel for this stuff.

What I'd be most concerned about is the time it takes to repair a pool or to run a scrub or SMART long test. As the drives get larger then the time it takes to do this tasks grows considerably. I migrated from six 2TB drives in a RAIDZ2 to four 6TB drives in a RAIDZ2. I gained a few more TB's of storage and reduced my drive count. This was not the cheaper option but it was the option I wanted. Now my scrubs and SMART long tests take considerably longer because I have two less drives in the stripe. Also the drives are different vendors so that adds something to the mix.

So what I'm saying is to state what you presently have and what you think you want to migrate to and I think someone could point out the pro's and con's to you, maybe with some specific details.