HPE Storage Users Group

So we are in the midst of migrating our old Netapp FAS3020 systems with Netapp disk over to our new Netapp v3140 which are diskless, they get their primary storage and boot from the 3PAR.

Interesting to know is that the Netapp multipath only permits 2 paths per LUN on an "active/passive" basis. The system auto balances primary paths for each LUN on different HBAs. As a fan of math and symmetry, this 2 path limitation challenged the feng shui of my SAN zoning model since each T800 has 4 nodes. I was able to satify my OCD by zoning each Netapp to both T800s, 2 paths to each, 4 per node, 8 total system wide, one path to each 3PAR node.



My storage plan is Filer A will build all it's aggregates from LUNs on T800-1, while Filer B does the same with T800-2. Since the Filers build raid arrays from the LUNs assigned, we opted for raid-0 since the 3PAR is already handling data protection (Raid 5+1 disk shelf survivability). For that reason, I prefer that all raid-0 arrays do not span storage frames to minimize points of failure.

Performance wise, the Netapp's appear to be very chatty on the fibre constantly streaming 300~400 iops of 8k blocks all day, every day. This is odd because I know the data that currently resides on the new hardware is typically idle and only accessed by a few IT users (we've halted all migrations until we can patch the Netapps). This has me suspicious of some a background disk scrub routine or other type of maintenence enabled on the Netapp that I need to discover and review. Charts below show trafic as monitored by 3Par System Reporter "High-Res" collection interval for the past 24 hours.

Constant 3% utilization of the LUNs assigned to the Netapp, a steady never ending stream of IOps. The spikes represent my snapshot cycles on the Filers, as a new snapshot is created the old 30 day old one is getting deleted/purged. Go WAFL? Hopefully I can locate and disable the cause of this. If not, I may move the Netapp to dedicated ports on the 3PAR hardware... currently I have them zoned to ports shared by Windows servers.

Hind sight being 20/20.... I have to report that the NETAPP is not thin provisioning friendly. Through the course of a week or so, it basically overwrites every block on the LUN. That said, do not waste your licensed thin provisioning capacity on NETAPP luns, fully provision them from CPG instead and save that metered capacity.

Hi Richard,
we have quite similar configuration using a Netapp v3170 cluster and EVA8400.
the main difference is that we use only 2 Ports on the Netapp to assign the Luns, and we use regular aggr's of raid group 8 built out of 25 luns sized 650g. All according to Netapp recommendation.
We are reaching to 10k iops for each netapp controller but the main problem is lack of real HA between the EVA controllers, whenever we need to upgrade the code it brings both controllers to restart which is long enough for the netapp to Panic.
The Good news is that we are moving to a 3PAR architecture, I will update on how this is going when we start moving on.

Hi Richard and all -

Did you discover a way to reduce or eliminate the fiber chattyness?

Thanks,
jc

I have a question about the whole architecture:
why using the 3PAR disks for the Netapp, Are they cheaper?
I assume that the T800 can serve more iops. but in terms of ROI, you will need to buy service & manage more array nodes.