#FAS2820 and DS460C agregate (local tier) creation.

1 messages · Page 1 of 1 (latest)

chilly plaza
#

Hello Netapps, we have a need for storage that will have two workloads.

  • VMware VM general workload on SSD and
  • NFS with for large number of images.

NFS will be used to store large number of images and videos that rarely will be accessed in small quantity, with possible write between 150-300GB images and video per day (continuously over 24h).

Planned system config:

  • FAS2820 - 12 x HDD 8.89TB disks
  • DS460C 20 x HDD 8.89TB disks and 10 x SSD 889GB disks.

What is confusing that different tools are creating different recommendations for aggregate creation:
A) ONTAP System Manager is recommending:
1 x HDD aggregate with 2 x RAID-TEC groups > 15 disks each = 12 data + 3 parity. and 2 spares.
1 x SSD aggregate with 1 RAID-DP group > 9 disks and 1 spare.

B) While NetApp Fusion sizing tool is recommending:
3 x HDD aggregate with 2 x RAID-TEC 8 disk group > 8 disk=5data + 3 parity and 2 spares. + 1 x RAID-TEC 10 disk group > 10 disk = 7 data + 3 parity and 2 spares.
1 x SSD aggregate with 1 RAID-DP 8 disk group > 8disk=5data + 3 parity and 2 spares.

Could anyone shed light if we need in this case 3 aggregates (pros and cons) if we have only 1 type of workload ? Is there any Pros if aggregate is split between physical chassis example 1 aggregate is on controllers and other is just on shelf , or it doesn't matter if it expands from controller to shelf?

lucid viper
#

We have a similar setup... we have split the disks, so that spinning disk aggregates are on one controller while the SSDs are on the other. We started with only the spinning disks in the system, which partitioned the disks for the root aggregates. We then assigned all disk partitiones to controller A and then added the SSD disks and created a new aggregate on controller B with the whole disks... In the "old days" ONTAP had an issue where if you had mixed performance disks the write cache could suffer if you pushed the slower disks... (it filled up the cache, so that the faster disks suffered)... I think they solved this (maybe by splitting the cache up for each aggregate, not sure... you may also want to look into tweaking the SSD aggregate so that it has inline dedupe/compression enabled because this is not automatically enabled on FAS systems...

chilly plaza
# lucid viper We have a similar setup... we have split the disks, so that spinning disk aggre...

@lucid viper Thank you for your reply and sharing your experience. WE already performed the same split of different type of disks, SSDs are on Controller A and HDD spinning disks are controller B. I will check regarding recommendation for FAS SSD inline dedupe/comporession config, thx for that. The main elephant in the room for us is how 32 x HDD on two different chassis should be split.

hollow violet
#

I can see doing this one if two ways:

one
use ADP and then assign all data partitions to one node and all SSD to the other. Create one aggregate with 29 drives (max raid size). Leaving three spares and make a 9-disk raid-4 or raid-dp aggregate

Two:
still using ADP. Create two aggregates of hdd, one on each node. In my casei works simply keep all drives partitioned and build 14 disks in each aggregate (leaving 2 spares per node). You could possibly increase to 15 disk leaving one share to get a little more capacity. Still have one node with all the SSD and one aggregate
Then use a flexgroup to allow ONTAP to distribute across both nodes

lucid viper
#

Not sure if the SAS loops are shared internally with the two ports on the back... but I think so so don't think it matters.... also this model isn't really geared for performance anyway... there seems to be three SAS ports on each controller, not sure what the 3rd is used for? Maybe to gain more throughput?

hollow violet
#

third is not used

#

If I recall, it is for an "internal" third connection

#

The 0b2 port is not used. It is disabled. If a cable is connected to it, an error message is generated.

chilly plaza
chilly plaza
lucid viper
#

Yes looks OK to me. The SAS disks are mainly partitioned because the two root aggregates have to be somewhere, so they will becore Root-Data-Data partitioned... And yes you can use a FlexGroup which essentially just creates multible volumes on different aggregates... It all depends on what you want, which works loads you have etc.. because there are also limitations to flexgroups... another way to go, would be to use FlexCache with which you can add the SSDs as a cache in front of the SAS aggregates... but you don't get the extra capacity of the SSDs but you get better performance... again dependant of your workloads... and streaming workloads does not make much sense to cache and I think FlashCache will automatically ignore sequential reads and writes as far as I remember (unless you set some special option that forces it to) but come to think of it, I think the FAS2820 already have a 1TB NVME internally that is used as FlashCache? It is kinda hidden from the GUI..

hollow violet
#

You can absolutely do 29 disks in a Raidgroup as long as you have Raid-tec. Partitioned or not. I have done this plenty of times. Fusion even allows it. Personally, I do something on the CLI as the GUI will NOT do what you want (not enough knobs).

aggregate create -node node-01 -maxraidsize 29 -raid-type raid-tec -diskcount 29

If you have partitioned drives, it will tell you that it will be paritioning 23 drives. If it fails (which I think it does), I just use the DIAG command to partition 2 drives to meet the minimum then try again.

#

I end up with a 229.62T aggr size in a single aggregate
I end up with 97.15T aggr size (194.3T total for two) using 14 drives per aggr
I end up with 105.98T aggr size (211.967T total for two) using 14 drives per aggr

outer bronze
#

If you pre-partition the NL-SAS disks, you might convince "aggregate auto-provision" (or GUI) to create the 29-disk-rg. But TMACs aggregate create command will certainly do the trick.

Another tip - for the non-full DS460C, for correct airflow, you must populate all disk slots closest to the front (4 disks x 5 drawers = 20). Because I like consistency and order, I'd probably put all NL-SAS disks there. The SSDs I would use in the "middle row left-most and right most" slots to create a "smile" in each of the drawers.

Like (X = empty, S=SSD, N=NL-SAS):
X X X X
S X X S
N N N N

chilly plaza
# hollow violet You can absolutely do 29 disks in a Raidgroup as long as you have Raid-tec. Part...

Thx for clarifying.
So Solution 1): One big aggregate is valid option. Only future expansion might be limited.

I wanted to test **Solution 2) **in ActiveIQ NetApp Fusion.
However, for some reason software is blocking mixing HDD Controllers disks with HDD DS460C shelf disks in one aggregate.
Fusion is allowing me only to create separate aggregate on Controller and on DS460C shelf. Example in screenshot.
Now is this a feature (valid recommendation) or a bug from software ? 🙄

image.png
chilly plaza
hollow violet
#

Yeah…option 2 does not work in fusion.

#

It’s a bug. I’ve been asking for a long time to get it fixed

#

Create aggregate
Set raid size to 14 or 15
Select 6 partitioned drives
Check the allow mixed box over on the right. Then select add drives
Select 8 or 9 whole drives

The bug is you can’t do a second aggregates because you only have 6 partitioned drives. But you can just double what you did

#

You can argue that you can just assign all the partitioned drives to one node then make an aggregate with 12 partitions and two-three whole drives (which still get partitioned) then a second aggregate of whole disks

Personally, I like symmetry when possible

#

When I get home I’ll try and send you my screenshot

chilly plaza
#

Thank you, I guess I have an option to trick it. I agree on symmetry. Take your time its a weekend 🙂

hollow violet
#

Wait. What? What’s a weekend?🤓