Pure Storage and NVMe
Storage controller-connected “NVMe NAND” and “NVMe over Fabrics (NVMe-oF)” are no longer just storage industry buzz words. Both technologies are here and primed to replace legacy media, the path to that media and the protocol that brings data to the application.
WWT and Pure Storage have teamed up to help you design next-generation data management solutions that perform and scale.
The good old days
It seems like yesterday when storage folks required extensive project plans to upgrade storage arrays, finalize data migration strategies or upgrade SAN fibre channel (FC) switch fabrics. After all, going from 2Gb/s to 8Gb/s FC enabled higher throughput capabilities for the storage arrays in the data center, which in turn improved application performance and the end-user experience.
Many in the industry remember the days when storage teams had to work extremely hard to gain every ounce of performance from spinning media to support mission critical transactional applications. The complicated process required downtime and needed extensive dependency mapping and troubleshooting to ensure optimal storage performance. Even after all this work, admins likely netted only a small increase in performance after the entire stack was considered.
8Gb SAN delivered great improvement at the time (circa 2008). But if the storage array only supported 4Gb/s or the host HBA only supported 2Gb/s, the end result could have been limited by upgrading only one piece of the puzzle. In legacy three-tier stacks, upgrading one component simply shifted the bottleneck from the fabric to the storage array or the host. If the array was CPU-bound or resource constrained, there was little that could be done to improve overall system performance outside of an upgrade or resource re-allocation operation.
Alternatively, if the array was disk constrained and overburdened in keeping up with host I/O requests, admins would look at different ways of spreading those logical resources (LUNs) out on the array to provide the most processing power to each block device presented. I/O-thirsty OLTP applications often forced customers to buy more disk capacity than they needed just to deliver the required performance. This process usually required downtime to re-architect the layout by using volume management software at the host.
Ah the good old days, right!?
Enter flash media
Thanks to next-generation flash NAND storage media, those days are long gone. Admins no longer spend days tuning spinning media on the storage array or laying out the LUNs to interact with the mission critical applications. Instead, in the era of all-flash storage arrays, engineers see the performance bottleneck shifting back to the storage network or even to the server requesting the I/O.
From my personal experience, I’ll never forget the first business day after a 12-hour maintenance window where we migrated all of our production database-backed storage to an all-flash storage array. I had a DBA ask me “Can you please slow the storage down? The host CPUs can’t keep up!” Ultimately, we had to add additional compute nodes to solve the problem.
In addition to the array performance, the legacy SCSI storage protocol, which has been the standard for decades, is far less efficient than newer protocols and creates increased demand on host-side CPUs. As storage media is modernized, engineers must also think about modernizing the networking path to that media from the server/host to gain even more efficiencies from all-flash arrays.
The storage industry is rapidly adopting NVMe protocol in array-based solutions. Just as flash NAND media provided advances over spinning disk, NVMe and NVMe-oF has improved the way we connect NAND media to the hosts that require extreme performance.
NVMe protocol replaces the traditional SCSI protocol, while NVMe-oF allows for much more parallel access from host to the storage media. The result is higher throughput, lower latency and decreased demand on host-side CPU.
There are myriad ways to accomplish this. Some OEMs are using traditional fibre channel to bring NVMe protocol from the host to the storage array. Others are using ethernet solutions like RDMA over Converged Ethernet (RoCEv2), NVMe/IB, NVMe over internet Wide Area RDMA Protocol (NVME/iWARP), or NVMe/TCP. Niche players are using proprietary drivers and cards to implement NVMe. It can get rather complex.
Customers often ask, “What are the benefits of adopting NVMe and NVMe-oF? What does this mean for my applications? How can we test this before implementing it in our infrastructure modernization plan?”
At the moment, our answer is simply that your mileage will vary as host operating systems continue to build new drivers for new protocols. At WWT, we’ve prepared our Advanced Technology Center (ATC) lab to help you test these solutions as they roll out into the industry and help find the right fit for your environment.
Introducing Pure Storage FlashArray//X90R2: DirectFlash Fabric
As we engage our customers around topics like NVMe, NVMe-oF and next-generation data center modernization, Pure Storage is frequently part of the discussion. An industry leader in the all-flash primary storage space, Pure is changing the game with innovations that help customers simplify their storage environment. Pure has an 86.6% Net Promoter Score — one of the highest in the tech industry and one that’s been backed up by our experience with their technology in the ATC.
Our ATC features a number of Pure's FlashArray™//X and //M products in a lab ecosystem that uniquely showcases storage integrated with other adjacent technologies. In the spring of 2019, WWT ATC engineers worked with Pure to deploy the newest and most powerful FlashArray in its portfolio — FlashArray//X90R2™. As Pure’s most performant storage offering to date, FlashArray//X90R2 includes Skylake CPUs from Intel, a DirectFlash™ NVMe modular design (DirectFlash modules and RoCE-connected DirectFlash storage expansion shelves), and stateless architecture coupled with the award-winning Purity™ operating environment.
Our testing on the Purity 5.X operating system, delivered as part of the FlashArray//X90R2, has demonstrated the value of the fine grain data reduction offerings Pure brings to the table. Our customers like the platform’s data density as well as its high performance, resiliency and efficiency.
FlashArray//X90R2 also delivers outstanding system performance by giving customers the option for NVMe-oF via RDMA over converged ethernet (RoCEv2). In our ATC, we can showcase how NVMe via RoCE compares to traditional iSCSI via 10GbE and, in turn, show how it compares to 16Gb FC. This type of comparison showcases the differences between each and can inform customers as to the right path to take relating to next-generation fabrics.
In addition to the FlashArray//X90R2, the Pure Storage footprint in our ATC consists of several additional FlashArray systems as well as FlashBlade™. We also work directly with Pure Storage engineering to beta test new software and hardware products before they’re released for general purchase.
WWT can also showcase new Pure Storage features and capabilities once released, such as:
- NVMe-oF DirectFlash Fabric™
- Active-Cluster Synchronous Replication™
- Flash-to-Flash-to-Cloud (FtFtC) data protection models
- Cloud Block Store™ — Purity in the public cloud
- Equinix data center integration into the ATC
One of WWT’s primary missions is to be the first to educate our customers. We invite you to see the latest innovations firsthand and test solutions before deploying them in your own environment. Our close relationship with Pure Storage and our experience with hands-on demos, sandbox environments and proofs of concept can help you simplify unknown elements before you invest.
ATC Flash Lab - Pure Storage Capabilities