Tuesday May 9, 2017, Blog 2: HCI Hype
Tremendous show, I’m constantly on the run! You're keeping me busy. I love it!
Reminder, Today is my Extension breakout session titled, "WAN Optimization for Storage Replication from SRDF to SyncIQ" at 3:00 PM Tuesday. The session focuses on Optimization using Connectrix between Data Centers. The MP-7840B can optimize replication between data centers for most IP storage replication: SRDF, RPA, DLm, Data Domain, SyncIQ, Data Mover...
I want to say a few words on HCI (Hyper Converged Infrastructure). I’ve visited many customers this past year and gathered feedback concerning their HCI experiences. I'll admit, I mostly visit large enterprises. Here’s what I’ve been hearing…
FC Remains the Benchmark
FC remains the undisputed benchmark for storage networking characterized by: mission critical, ultrahigh-speed I/O, ultralow response time, extremely available, secure and relatively simple to operate. FC is offerred with comprehensive visibility tools, monitoring, diagnostics, reporting and management. This applies to distributed systems, virtualized environments, mainframe and replication environments. Every other form of storage networking makes compromises lowering the level of service compared to the FC benchmark.
If there’s no better storage network than FC, then FC could be applied to ANY application with no viability concern. In other words, you know it’s going to work in every case. This will be true no matter how far applications evolve. Granted, a SAN may be overkill for some low tier applications, I admit that. Nonetheless, some of the HCI issues and aggravation experienced by customers seem to negate some of the benefits expected by deploying their new, different and unknown storage network.
I think people have forgotten the true virtues of FC + array. A deterministic reliable high-speed network deployed and operated by the very storage folks responsible for their own SLAs. Predominantly, I hear that a Brocade SAN is a “set it and forget it” network. There’s no better storage network!
Gartner Hype Cycle
HCI is currently at the Peak of Inflated Expectations! If you don’t have Gartner’s report, you should think about buying it. So, what does it mean to be at the “Peak of Inflated Expectations”? It means that you’re about to drop into the “Trough of Disillusionment” and that’s exactly how it sounds. Can HCI live up to the extensive hype of the marketing machines set forth to conquer your hearts and minds? According to the customers I’ve talked with, the answer is an overwhelming - No! Hence, the reality trough. Following are some reasons why HCI does not live up to expectations…
If you deploy HCI software to a considerable number of servers, then distributing your data across those servers to buildout your HCI… at this point you’re essentially locked in. Who’s kidding who? When you want to change HCI technologies, how is that migration going to work? Can you change just part of your HCI deployment during a migration, or is it more of an all or nothing deal?
What about FC? Can you change technologies when using a SAN? Yes, of course because there's no lock in. Customers switch from Cisco to Brocade all the time (LoL, sorry I just couldn’t resist that one! It does happen to be true though!). A FC network and its management don’t create a technology lock-in. HBAs are not a lock in. The FC network is not a lock in. The storage arrays are not a lock in and the servers are not a lock in. Migrations are not all or nothing, and they can occur over a period of time bit by bit (excuse the pun). The various components from initiator to target all provide choice and the ability to relatively easily make changes if desired.
Can HCI scale like FC? Clearly, the answer is no. FC and storage arrays are HIGHLY scalable, meeting the needs of nearly every environment from minute to massive.
With HCI, the number of nodes feasible within a cluster is limited, after which, you build more clusters ($$$). Nodes within a cluster share volumes and these volumes have redundancy in the event of failure. Each volume can accommodate only so many IOPS, and pushing a cluster to scale will overwhelm not only the volumes but also the shared Ethernet network connecting them, ultimately creating bottlenecks. All the while, the servers are processing I/O and de-encapsulating/encapsulating data onto the network. Also consider, CPU is not a limitless resource.
HCI by definition is aggregated compute and storage. Adding compute, adds storage whether you need it or not. Adding storage, adds compute whether you need it or not. This can produce significant inefficiencies that don’t exist in modern storage array infrastructures, i.e., thin provisioning.
FC is incredibly fast and frames are switched in nanoseconds. Undoubtedly, HCI will have worse response time performance.
IOPS (I/O Per Second). The number of IOPS that can be accommodated will be the greatest on servers with an HBA, a FC network and a storage array. HCI will have significantly worse IOPS performance.
Is an RPO of 5 minutes or more adequate? Depending on the remote location plus the speed and quality of the WAN, a sufficient RPO and RTO for HCI may be a problem. Array based asynchronous replication across extension provides RPOs measured in single digit seconds.
How about high availability using multiple lossless connections? Is HCI replication traffic compressed and encrypted? Would that cost extra in terms of licenses or CPU cycles? Keep in mind, it is prudent to encrypt all data leaving the secure confines of your data center.
As you can see, we’re currently at the peak of the hype cycle and about to fall deeply into the disillusion reality of HCI. Don’t get caught up in the hype. Moreover, who’s already at the “Plateau of Productivity”? Brocade Fibre Channel and Extension, that’s who!
Be Good! Have fun tonight!
Mark Detrick @Extension_Guru
Brocade Principal Solutions Architect#HCIFCHype#BrocadeFibreChannelNetworkingCommunity