Log in

No account? Create an account

Josh-D. S. Davis

Xaminmo / Omnimax / Max Omni / Mad Scientist / Midnight Shadow / Radiation Master

Previous Entry Share Next Entry
Why, NetApp, Why?
Josh 201604 KWP
Why do NetApp specialists like to put 50+ spindles into one RAID-DP aggregate?

Customers are defensive of this because "this is what (NetApp|Our NetApp Specialist) recommends. One big aggregate."

RAID-DP is not much different from RAID6. RAID6 is not much different from RAID5 except for the parity quotient.

Having too many spindles in one array effectively is both read and write amplifying.

Also, heavy random workloads (databases) should be separated from everything else (apps, OS, Logs).

To their defense, I don't know what their raid group size is; however, my current customer gets 30MB/sec to the OS on a disk to disk copy. During backups of production boxes, my NIM server's disk-to-disk copy speed drops down to 1MB/sec. Peak is 50MB/sec for a snap/clone.

By comparison, I can get 60MB/sec out of five SATA disks in a RAID6 on a Pentium 4 Linux box. A midrange array with 50 SAS/FC disks should get many more times that.

I've run into this issue at a few customer sites. I'm not sure if it's due to adding spindles, or planning, or something else. I just know that it seems that multiple customers are being robbed of performance and the default configs seem to be suspect.

  • 1
I'd hazard a guess that it's a holdover from Oracle telling people that more spindles in an array means less iowait when writing transactions to the DB.

Hrm... Maybe...

Which would still make sense in a RAID-10... :/

  • 1