Introducing the Pure Storage FlashArray //XL
In this blog
Pure Storage released their newest model in the unified storage family on December 8th called the FlashArray //XL (code name Oxygen). This new hardware model brings a redesign to the chassis form factor but fully supports all the same features and functionality of the Purity FA operating system that we know and love today. Let's dig into the hardware changes and I'll give you my thoughts on what this means for both data center and storage teams.
The new FlashArray //XL comes in two sizes and top out the FlashArray lineup by adding both scale and performance, while still maintaining a small datacenter footprint as compared to their competition. The //XL130 supports up to 968TB of raw capacity and the //XL170 supports up to 1.4PB of raw capacity.
You are probably saying to yourself, "is it just a bigger and faster FlashArray?", and you would be partially right. At first glance, we can see that the maximum raw capacities (before their always-on, best in class data efficiencies) for the new //XL models are higher than the other //X models in the lineup, but we need to look under the covers to see all the other changes and updates that Pure has released in the new //XL line.
If you are familiar with the hardware for the current FlashArray models, you know that the current chassis form factor used for both the storage controllers (w/ disks & cache) or any of the add on DirectFlash shelves are 3U in size have 28 drive slots and support dual power supplies.
For the new //XL models, Pure has moved to a slightly larger 5U chassis design supporting four power supplies that operate in a N+2 configuration. If any two of the power supplies have power, the chassis will be online and will serving workloads. A larger chassis form factor also allows for more drive slots so instead of the controller chassis supporting up to 20 drive slots for Direct Flash Modules (DFM) and Direct Memory Modules (DMM) along with two or four NVRAM modules. The new 5U chassis now supports 40 drive slots for both the DFM/DFMD/DMM storage. This equates to a 20% more efficiency savings for Rack Density in the data center while providing up to 68% more capacity as compared to the //X90. As your capacity needs increase, you can add on the same DirectFlash Shelves (with DFMs) that the //X models use to meet your storage needs. More to come on the new DFMD storage below.
As mentioned above, the current FlashArray //X models include two or four dedicated slots for NVRAM modules. The NVMRAM modules are dual connected to each stateless controller and provides super-fast write acknowledgements back to the hosts. The new FlashArray //XL models no longer use dedicated NVRAM modules, instead Pure has distributed the NVRAM by building it into their new Direct Flash Modules with Distributed NVRAM (DFMD) media. The DFMD not only includes the NAND storage found in the DFMs, but also includes a small amount of NVRAM in each DFMD. So instead of having two or four dedicated NVRAM modules per system, now you will have 20 (for //XL130) or 30 (for //XL170) DFMDs per system along with the number of DFMs to meet your capacity needs. Note, there are no dedicated DFMD slots in the //XL series controller chassis so you can insert DFMDs and DFMs into any open slots.
Intel's latest Ice Lake CPU line not only supports more memory channels and faster memory, but also supports PCIe Gen 4 with more lanes per CPU which equates to more bandwidth for connectivity. The new FlashArray //XL models support double the IO ports as compared to the FlashArray //X models and by removing the use of mezzanine or LOM slots, each controller provides 9 PCIe Gen 4.0 slots for connectivity.
Just pick the speed, quantity and type of HBA cards for connecting to another FlashArray for replication, connecting additional Direct Flash Shelves for more capacity, or connecting to hosts to provide storage via FC/iSCSI/ethernet. And with support for 64Gb FC (32Gb supported at release) or 100Gb ethernet, the FlashArray //XL is ready for tomorrow's workloads and IO requirements.
In the current FlashArray //X (and //C lines), High-Availability (HA) is done via an internal Non-Transparent Bridge (NTB) between the controllers. NTB allows the stateless controllers to share the same back-end disks by sharing part of their memory with the other controller via a high-speed PCI express link. This is also what allows all the Host IO ports to be used on both controllers even though one of them is primary and the other is secondary – waiting to take over in the event of a Non-Disruptive Upgrade (NDU) or a system failure on the primary controller. Since the //XL models support more IO ports and bandwidth, the NTB is no longer fast enough to keep up, enter 100GbE RDMA over Converged Ethernet (RoCE) cross-controller links. The new RoCE cross-controller links replace the NTB used on the //X and //C models and provides the necessary speed and throughput needed for these larger systems under max loads and during failover/upgrades.
If you are familiar with Pure Storage products you know that NDUs have been available since the beginning and have allowed for upgrading the controllers (without needing to migrate data) from the original FA line released in 2012 to the //M and //X models available today.
At launch, a NDU from the //XL130 -> //XL170 is fully supported. With the change in chassis form factor between the //X to a //XL (just like from a FA to a //M), a few more steps will be required for the NDU. This will be available and supported in the second half of 2022, although the use of ActiveCluster is fully supported today between a FlashArray //X and //XL and would allow for non-disruptively moving live workloads and data between the FlashArray models.
A new larger hardware design with more IO ports and support for more flash drives must equate to more performance, right? Absolutely. Initial reports are showing up to a ~65% improvement for sustained throughput compared to the //X90R3 for datasets with and without data reduction at different block sizes. They are also seeing up to a ~70% increase in peak IOPS on the new models and with the additional IO ports, throughput has increased from 20GB/s up to 36GB/s on the //XL170 all while maintaining the same read latencies that are available on the //X90 today. Direct Memory Modules (DMM) are still supported on the //XL series and come in either a 4-pack or 8-pack. Using DMMs for cache specific workloads can reduce read latency times from 250μs down to as low as 150μs.
As you can see, there are quite a few new enhancements that come with the rearchitected design for the FlashArray //XL. The best part is that all the same reliability, high-availability, security, data efficiencies and features that the Purity FA operating system supports today are available day 1 on the new FlashArray //XL models and between //XL and //X or //C models for services like Replication and DR. With a software update planned for 2022 we expect to see both the Volume and Snapshot counts to be increased to support up to 32k volumes and 256k snapshots per system which will give the storage teams what they need to stay ahead of their customer requests. And for the data center teams, we know that datacenter space is becoming more expensive and harder to find. Imagine being able to provide 5PB of effective capacity in 11U of rack space. Compare that to 1+ full racks of storage that the competitors require for a similar sized solution and you can quickly see the savings and efficiency of the new FlashArray //XL.
In the end, I think Pure has done it again and delivered exactly what customers were asking for plus so much more with this new FlashArray //XL release. By redesigning the chassis to not only allow for higher capacity DFMs when they are released but also increasing the drive slot density in the controllers, improving HA and resiliency. I'm excited to help our customers put the FlashArray //XL to the test and to see how solves their storage challenges moving forward.
Have questions about Pure Storage and the new FlashArray //XL? Our subject matter experts are here to help. Contact your WWT account team or reach out to us directly if you are interested in testing out the new FlashArray //XL or would like to partake in one of our Primary Storage workshops where we help our customers see how a FlashArray //XL may fit into your organization's data center storage design.