Root Volume On SATA or SAS Disks? Or Does it Matter.

[cgeck0000]

I am working on a design for a 2240HA-4 system with internal SATA disks and an external disk shelf with SAS 15K drives.  I would like to know if there is a recommendation as to where the root volume for both controllers should live.

Is it best to have the root on the internal SATA disks or the external SAS disks?  Would it matter performance wise?  Would things be easier one way or the other when time comes to move up from the 2240 to say a 3240?

[scottgelb]

This could spark some good debate :). It won't make a difference on a head swap later. Performance isn't something we see an issue with on root as with raid_dp resilience is there too on sata. Cost per drive could be a consideration.

I do like to mirror root between aggrs. So if root is on sata, create root_bkup or similar name and snapmirror every hour or whatever makes sense. Then if the root aggr fails, you can specify the other aggr as root then it creates AUTOROOT. You can then break the mirror then mark as root and reboot again. Could be overkill for most but similar to cluster-mode where having a local and remote mroot mirror on every node is best practice.

Still didn't answer your question exactly. But some also put all sata on node1 and all sad on node2 of a cluster then the decision for root is easy with 1 aggr. If you mix sata and sas on a node some might lean toward sas for root...I would but could argue the Sata case too.

[cgeck0000]

So it sounds like in general the root volume can reside where ever without issue to the system itself and all functionality.  I like the sound of the idea of putting all SATA on one node and all SAS on another, because it would give you more useable storage per node. I always thought you had to use similar storage on each node for the root, but I guess not.

Thank you for the insight Scott

[scottgelb]

Each node can run with separate disks so makes sense…can also help performance since in edge cases SATA CPs can slow down SAS when shared on a node.

[cgeck0000]

Are CP flushes regulated by each node depending on what disks they are destine for?  Or does the node flush all data at once no matter where it is going?  I've been looking for NetApp documentation with more detailed information about the inner workings of the disks and aggregates and data writing pertaining to performance.  Please point me to anything like that.  Thanks.

[scottgelb]

Some tribal knowledge on this and some support.netapp.com mentions of it… pretty much every 10 seconds or half full NVRAM, ONTAP flushes writes in RAM to disk. So if SATA takes longer to Consistency Point that could slow SAS on a busy system… but I do remember reading something about partial CPs to help alleviate this potential edge case.

[shawnodell]

best practice is to dedicate an aggregate to the root.  This is if you have an aggregate corruption later, you can get the system online much faster.  Not sure if you can spare 3 disks, would be cheaper to use sata...  I had noticed a bug with ACP causing excessive I/O on the root volume, disabled ACP at the time, not terribly interested in turning it back on.  On root volumes that were on sata this was further exacerbated.  Otherwise I haven't seen to much I/O hitting the root volume, I'd say go ahead with sata.   I've had a 2040 I have in a branch office on the external sata for a while now.. no problems (also it's partner on the internal sas...)  no problems of note...