Maximizing Cisco 32Gbps Fibre Channel Infrastructure
As Cisco rolls out a new line of cost optimized MDS 32Gbps Fibre Channel switches, expect to see increased adoption of transaction and analytic workloads, a milestone for active data applications and services in hyper-scale enterprise and service provider segments. This, along with mainstream adoption of NVMe flash storage are all good signs for driving throughput, low latency applications that are pushing active data deeper within the Enterprise on 32Gbps Fibre Channel infrastructure. Timing of this launch is excellent because the oncoming tsunami of data – being driven by IoT, Augmented Reality and Machine Learning – will continue to stress infrastructure teams to manage this unprecedented scale while trying to maintain budgets, performance and SLAs.
Mainstream deployment of Cisco’s ultra-fast MDS 32Gbps FC switches also enables the active data workloads that will benefit from the throughput and CPU power that Cisco is providing with these new switches and UCS servers. This compute infrastructure has the I/O horsepower to “fill the pipe” – BUT only if the rest of the infrastructure can sustain these increased performance levels.
Considering that the latest generation of multi-core scalable Intel processors can easily sustain a million IOPs and up to 25GBps of throughput, but when performance stalls it’s clear that there is another choke-point.
You guessed it – storage.
Existing (legacy?) All-Flash Arrays (AFAs) can only provide about 3-5Gbps of throughput at about a million IOPs with latencies pushing well beyond 1000uS. The outcome is those brand new multi-core servers and 32Gbps FC switches become “data starved”. This is the inconvenient truth of the current class of AFAs on the market.
To double-click, take a look at the impact that storage bottlenecks might have on an Oracle RAC deployment. As congestion occurs, I/O is queued and the throughput drops to a level where users and apps begin to experience pain – in the form of transactional delays. These roadblocks become cumulative and can snowball to the point where the infrastructure admins must take action. The first reaction may be to re-architect the database – distribute the load to avoid these bottlenecks. This can be a band-aid that typically involves procuring more database licenses and infrastructure and you guessed it – storage. This is even though the licenses, CPU cores and FLASH capacity are being under-utilized, there simply isn’t sufficient I/O processing power to scale. Until the architecture issues are addressed, they’re merely kicking the bottleneck down the road.
Solving this requires a completely new approach that eliminates I/O queuing to take full advantage of the performance of Cisco UCS compute and MDS switching infrastructure. The outcome is to accelerate the critical applications that have been throttled, providing seamless scaling and growth to support mixed workloads that scale in multiple dimensions.
This succinctly describes Vexata’s mission to streamline how enterprises unlock the core value of active data by accelerating the applications and insights that drive digital transformation. Vexata Scalable Storage Systems scale up to 64 NVMe or Intel Optane (3D XPoint) SSDs across a scale-out architecture that connects to the Cisco MDS via sixteen 32Gbps FC ports.
Vexata has worked very closely with Cisco to certify interoperability of the VX-100 Scalable Storage Systems with the MDS family of Fibre Channel Switches. This interoperability gives customers the confidence that the solution will perform across a broad range of application workloads from heavy transactional systems where IOPs are critical as well as analytic systems that place excessive system throughput loading. Vexata is the only system available that can simultaneously deliver in excess of 60GB/s of throughput and over 7 million IOPs and fully utilize 32Gbps FC links at scale.
Typically, infrastructure architects need to make tradeoffs between transactional or analytic workloads, but the combination of Vexata and Cisco 32Gbps Fibre Channel switches enable simultaneously support of hybrid transactional and analytic processing (HTAP) workloads.
HTAP is typically a pressure test for data infrastructure solutions, but as referenced in a recent ESG Lab validation report, Vexata Scalable Storage Systems can handle Oracle HTAP with performance to spare.
To learn more about how Vexata accelerates Oracle HTAP workloads, view this webcast:
NVMe & Oracle Database: Making Hybrid Transaction/Analytical Processing Possible.