Category Archives: Hardware

ADP(v1) and ADPv2 in a nutshell, it’s delicious!

Ever since clustered Data ONTAP went mainstream over 7-Mode, the dedicated root aggregate tax has been a bone of contention for many, especially for those entry-level systems with internal drives. Can you imagine buying a brand new FAS2220 or FAS2520 and being told that not only are you going to lose two drives as spares, but also another six to your root aggregates? This effectively left you with four drives for your data aggregate, two of which would be devoted to parity. I don’t think so. Now, this is a bit of an extreme example that was seldom deployed. Hopefully you had a deployment engineer who cared about the end result and would use RAID-4 for the root aggregates and maybe not even assign a spare to one controller, giving you seven whole disks for your active-passive deployment. Still, this was kind of a shaft. In a 24-disk system deployed active-active, you’d likely get something like this:

Traditional cDOT

Enter ADP.

In the first version of ADP introduced in version 8.3, clustered Data ONTAP gained the ability to partition drives on systems with internal drives as well as the first two shelves of drives on All Flash FAS systems. What this meant was the dedicated root aggregate tax got a little less painful. In this first version of ADP, clustered Data ONTAP carved each disk into two partitions: a small one for the root aggregates and a larger one for the data aggregate(s). This was referred to as root-data or R-D partitioning. The smaller partition’s size depended on how many drives existed. You could technically buy a system with fewer than 12 drives, but the ADP R-D minimum was eight drives. By default, both partitions on a disk were owned by the same controller, splitting overall disk ownership in half.

8.3 ADP, R-D

 

You could change this with some advanced command-line trickery to still build active-passive systems and gain two more drive partitions’ worth of data. Since you were likely only building one large aggregate on your system, you could also accomplish this in System Setup if you told it to create one large pool. This satisfied the masses for a while, but then those crafty engineers over at NetApp came up with something better.

Enter ADPv2.

Starting with ONTAP 9, not only did ONTAP get a name change (7-Mode hasn’t been an option since version 8.2.3), but it also gained ADPv2 which carves the aforementioned data partition in half, or R-D2 (Root-Data,Data) sharing for SSDs. Take note of the aforementioned SSDs there, as spinning disks aren’t eligible for this secondary partitioning. In this new version, you get one drive back that you would have allocated to be a spare, and you also get two of the parity drives back, lessening the pain of the RAID tax. With a minimum requirement of eight drives and a maximum of 48, here are the three main scenarios for this type of partitioning.

12 Drives:

ADPv2, R-D2 ½ shelf

24 Drives:

ADPv2, R-D2 1 shelf

48 Drives:

ADPv2, R-D2 2 shelves

As you can see, this is a far more efficient way of allocating your storage that yields up to ~17% more usable space on your precious SSDs.

So that’s ADP and ADPv2 in a nutshell—a change for the better. Interestingly enough, the ability to partition disks has lead to a radical change in the FlashPool world called “Storage Pools,” but that’s a topic for another day.

NetApp refreshes entire line of FAS and AFF platforms

Today NetApp announced a complete revamping of both the FAS and AFF lines and with it a divergence in model numbers. My favourite improvement is that NetApp has changed the way FlashCache gets delivered; now all FAS platforms can take advantage of FlashCache using an M.2 NVMe device onboard the controller, even the entry-level models; in fact, it’s standard on all models. In the realm of connectivity, both the top-end FAS as well as all AFFs can now offer not only 40GbE but 32Gb FC as well, first to market for both of these.

Without further ado, here are the new models in the FAS line:

  • FAS2620 and FAS2650
    • Appears to be the same 2RU enclosure as the FAS2240-2, FAS2552, and DS2246, likely with an upgraded mid-plane.
    • FAS2620 holds 12 large form factor (3.5″ NL-SAS/SSD) drives internally
    • FAS2650 holds 24 small form factor (2.5″ SAS/SSD) drives internally
    • Both models come with 1TB of FlashCache
  • FAS8200
    • Appears to be the same 3RU enclosure as the FAS8020
    • 1TB of FlashCache is now standard, upgradeable to 4TB
  • FAS9000
    • This all-new chassis separates the I/O from the controller so there are no more onboard ports and all I/O is done using PCIe cards, 10 slots per node.
    • 2TB of FlashCache are now part of the standard configuration, upgradeable to 16TB.

And the new AFF line now consists of:

  • A300 (Same chassis as FAS8200)
  • A700 (Same chassis as FAS9000)

Strictly the numbers*:

Model RU RAM NVRAM (NVMEM) Max HDD (SDD) Max Flash Cache Max Flash Pool Onboard UTA2 Onboard 10GbE Onboard 10GbE Base-T Onboard 12GB SAS PCIe Expansion Slots Cores
FAS
FAS2620 2 64GB (8GB) 144 1TB 24TB 8 4 4 N/A N/A 12
FAS2650 2 64GB (8GB) 144 1TB 24TB 8 4 4 N/A N/A 12
FAS8200 3 256GB 16GB 480 4TB 48TB 8 4 4 4 4 32
FAS9000 8 1024GB 64GB 1440 (480) 16TB 144TB N/A N/A N/A N/A 20 72
AFF
A300 3 256GB** 16GB 384 N/A N/A 8 4 4 4 4 32
A700 8 1024GB 64GB 480 N/A N/A N/A N/A N/A N/A 20 72
  • *Numbers are per HA pair
  • **16GB Carved out for NVLOGS

Performance Improvements

The FAS2600 comes with 3 times as many cores, twice as much memory and >3 times the NVMEM than that of the FAS2500 and brings 12Gb SAS and 1TB of NVMe FlashCache is expected to perform 200% faster than its predecessor running 8.3.x, making the entry-level line of controllers smoking fast. The 8200 has twice as many cores and four times as much memory as the FAS8040 and also comes with 12Gb SAS and 1TB of FlashCache, making it roughly 50% faster.

The new top-end model, the FAS9000 goes modular, decoupling I/O from the controllers. This performance monster which has 2TB of FlashCache standard and 20 PCIe slots for I/O is expected to run 50% faster than the FAS8080 on 8.3.x. A cluster of 24 FAS9000 nodes (12 HA pairs) scales up to as much as 172PB.

FAS9000 AFF A700 Chassis

Here’s how the new models map to the old:

New FAS platforms

As for the new AFF models, the A300 should get about 50% more throughput than AFF8040 running 8.3.1 while the A700 aims to replace the dual chassis AFF8080, saving four precious rack units but still providing 100% more IOPS, in fact it should be able to handle about double the workload at half the latency.

Oracle testing

And here’s how the new AFFs line up with the existing ones:

AFF model alignmentThe new lineup, both FAS and AFF are definitely addressing some concerns; FlashCache not only available throughout the FAS line but standard as well is a move in the right direction as is the addition of 12Gb SAS. The introduction of both 40GbE and 32Gb FC into the mid-range and upper models of both lines should provide the fire hose required to deliver all that new controller and storage back-end performance. The two new AFF model numbers lead me to believe that they may be leaving room in the middle to add models to the line.

While ADP has been around for a while and is a great work around to dedicated root aggregates, I would love to see NetApp move away from root aggregates completely and do something with M.2. I’ll keep my fingers crossed for this one, but won’t hold my breath either.