#which C series should we use to replace FAS8200?

1 messages · Page 1 of 1 (latest)

haughty wharf
#

We are thinking to use a HA pair of either C250, C400, or C800 to replace HA of FAS8200. What guidelines are when we consider which C platform should we use?
For instance, should we collect the average of IOPS, Throughputs on currect FAS8200, then to find out which C platform can provide the same level of IOPS and throughputs?

thin bobcat
#

The C-Series is very much underrated... at first I think NetApp was trying to position it as a backup target to replace spinning media... but the fact is that unless you need very low latency (and since you are comming from a FAS I guess yuo are not), the C-Series is a big upgrade in performance. We have installed many C250 and C800 as primary systems with customer with 100-500 VMs. I know it doesn't say much, because you can have just one VM that requires very low latency which might push you towarads the A-Series... By far the C800 make the most sense simple because it can have the 48 drives installed and the small footprint. But yes I would suggest installing tools like Nabox to collect performance data over a month, and from that your local NetApp pushers should be able to find a suitable replacement.. (Dependant of which disks you have installed in your FAS8200, my guess is that the C250 will outperforme it with just 8 disks installed) 🙂

wanton sluice
#

We migrated from a FAS8200 with 48 4TB NLSAS to a C250 with 8 disks, since 8x15TB the minimal configuration is, this was more then enough for that use case

north copper
#

Probably a C250 or maybe a C400 if the 8200 ran hot.
You can work with your account team to model the system in our fusion tool to see how it looks on each platform.

nimble orbit
#

I would also ask for a quote of a C800 surcharge compared to C400 is quite low but the additional memory (over 1.2TB per node) helps for read caching (reads only have higher latency when they come from disk).

wanton sluice
#

And with a C800 you already have 48 disk slots compared to at least one external shelf for the c400

dull fjord
#

The price jump from C400 > C800 is a LOT less than you might expect, and some have made the argument that the juice was worth the squeeze to skip up to it based on the headroom alone. Don’t sleep on the C800 thinking “oh I don’t need that much box.”

terse storm
#

What @dull fjord said! We’ve sold a lot of c800 units just because the price delta was usually inconsequential

wanton sluice
#

Personally I would like to see C90/C70 🙈

terse storm
#

I think many of us really would love to see that

umbral cipher
#

I would be surprised if those were not upcoming soon (maybe under a different name though...? ) Possibly something to announce at Insight? Who knows....

rain seal
#

In the Netherlands we skip the C400. C250 for small configs, but if possible always a C800! Not only for backup/SM/SV, but even production workloads for mixed (Cifs VMware) enviroments including MC-IP.

weak smelt
wanton sluice
#

We will now order a C800 instead of a 400er, price difference is like +/- 15%

terse storm
#

Big warning on the x800 platform:
X1146 does both optics and twinax, but no RoCE
X1148 does ONLY twinax and RoCE
So if you are looking at nfs/rdma , requires RoCE and the x1148

Most other platforms only support the x1148 with both optics and twinax.

The x800 with that card gets too hot using optics (the Mellanox card)

#

I’ve had a few not realize this and order the wrong card

wanton sluice
terse storm
#

I didn’t go there but…
ONTAP supports only MPO fiber. However, I’ve been told that you can talk to your sales team and request the feature policy variation request (fPVR) to use the Cisco BiDI optical transceivers (which is the LC fiber). So it works, but unless you have the feature pvr it will but be supported.

#

With that said I have not had a customer use this fPVR yet. I do not know the ins and outs of supportability (which cards, which platforms, etc)

vocal salmon
#

The FAS8200 is a mid-range system as is the C400. As has been noted, the C-Series performs very well, accounting for the higher latency (2-4 msec) and the pricing is very disruptive to competitive offerings.

nocturne mesa
#

Think of it as a budget AFF in terms of performance. It's slightly more latent, but most users won't notice unless you have a SSD only backend before migration.

wooden ember
#

We just replaced A700 with C800, although the CPU is quite old, the system is really fast and takes up only 4RUs... Would have loved to directly upgrade to C70 though

nimble orbit
#

Yeah, C800 is awesome. But I don't actually think they will announce C70, C90, C1K yet.
The current C-series models are simple "enough" at the moment. And you already have a bit more latency because of the QLC-SSDs so I don't think you would gain much latency-wise by adding the new Sapphire Rapids CPU to enable QAT. Not sure if hardware-offloading the compression would really help much since you already have the bottleneck with the SSDs.
At least you would have much more headroom with the new CPUs so higher maximum IOPS.

wooden ember
#

It's only the DR system, so load and performance are not that important

wanton sluice
dull fjord
nimble orbit
#

Sooo you confirm new C-Series models at Insight? 😉

umbral cipher
#

apparently they won't be called C70 though 🤷‍♂️

dull fjord
nimble orbit
wooden ember
#

So this is what I did today: svm-dr to C800... As you can see, QLC flash is fast enough, 2.08% Disk busy while doing 4gbyte/sec writes... Need more network bandwidth.... And maybe then I could even saturate A70 CPUs...

grafik-2.png
limber ferry
#

easiest non-disrupt process to migrate from FAS8200 to C250 ? deploy the C250 into existing cluster and then copy\snap data ? then remove FAS8200 nodes\diskshelves from this cluster ? thanks

terse storm
#

Svm migrate

#

Get to a version that supports both platforms.

#

Does not migrate luns. For VMware datastores it’s always easier to create a new svm and do storage vmotion

umbral cipher
#

Or join them into a single cluster and use vol move to move the data over

terse storm
#

Unless space isn’t available

nocturne mesa
#

SVM DR would be better if you can do a cutover, but both methods are valid (join cluster +vol move + remove old nodes/SVM DR)

terse storm
#

SvmDr is faster/easier if it’s supported with your nodes

limber ferry
#

the new C250 is being installed a few racks away so we would just need an IC connection for this SVM migration ? or LIF ? thanks

dull fjord
terse storm
#

All nodes need inter cluster lifs. It’s a full mesh network. Every node needs to be able to talk to every node using inter cluster lifs. That could be as simple as creating a new non routable vlan that all nodes have access to. As the ports (vlans) to each node. Create lifs on the new vlan and then create the cluster peering

wooden ember
#

if you have created a 4 node cluster with cluster switches, you will have cluster lifs (where the 4 nodes talk to each other) and intercluster lifs (where the svm-dr data is delivered from the primary system)

terse storm
#

To be clear:
Cluster LIFS are dedicated to local INTRA-cluster traffic
INTER-cluster LIFS are for communications between clusters

#

In either case, every node will have two CLUSTER lifs. Every node should have at least one INTER-cluster lif of you pan on any replication

limber ferry
#

we'll just have the 2 new C250 nodes when done....our CN1610 switches are EOM\EOL and we need to be switchless (can only have 2 nodes for that config)

terse storm
#

And the x250 platform cannot connect to those switches (cn1610). (So they can, it’s not supported and you give up two data ports to make it work, won’t go through it here

viral geode
#

I'm a big fan of the C800... Performance, compact, ports, expansion, and if you work well with your local netapp rep; the price delta can be negligible... Opens up this very versatile platform for even more workloads later.

fast lion
#

I'm running a DR Nas that's 2 8700 pairs and an ha pair of c800's.
8700's do big stupid workloads that don't compact further.
C800's hand everything else (1200 vols between them).

Prod is the same but an A800 instead of c800.

haughty wharf
#

@terse storm
Svm migrate Get to a version that supports both platforms. Does not migrate luns. For VMware datastores it’s always easier to create a new svm and do storage vmotion

Have some questions on your comments:

"svm migrate" works if I wanted to migrate data from a cluster to the other, correct?

In our case, C800 will be in the same cluster with FAS8200 which I wanted to replace. I can use either "vol move" or "storage vmotion" in VMware for migrating datastores once C800 joined into the cluster. How would you compare these two methods? Are there any significant benefits by using one over the other?

terse storm
#

Migrate/svm-dr only works between two different clusters , not within the same cluster

For volumes not hosting datastores, a vol move works just fine.

For volumes hosting nfs datastores, I find it better to actually do the storage vmotion in order to get the best possible efficiency update especially when going from FAS to AFF.

If you have any luns, there is a way to move those also. Can touch on details if there are any luns in play. Hopefully that helps a bit

#

The down side to the storage vmotion is that you will lose any volume snapshots!

haughty wharf
#

Thank you and all others for your providing all infor.

Currently, we are using 4x10G between FAS8200 and NX 3132q-v as intra cluster connections.
What options does C800 provide for using intra connections.?
Can we continue to use the 4x10G connections or has to be 40G?

Also, for customer connections, there are currently mixed 100/40/25 GbE on other nodes. What is your recommendation on what throughput we should use on the C800?

terse storm
#

You really need to check hwu.netapp.com and verify supported connections for each platform. I’m really not sure if the x800 platform even supports 10g. At a minimum it will support 40g using regular Cisco-coded twinax cables to the nexus 3132 switches.

#

Just make sure everything falls into the supported category meaning be sure to update the nexus code to current supported along with the rcf. I think the current is 1.12a

haughty wharf
terse storm
#

The c800 supports apparently all speeds. For 10/25 you need an extra nic

For 40 or 100 you just use the onboard

Go to hwu.netapp.com
Select c series
Select c800
Select ONTAP 9.14
Search
Select “supported connections” and see

haughty wharf
#

Are there restrictions on only a certain ports can be used for interconnections, assuming ports speed are all the same between them?