Discussions of enterprise storage networking often focus too much on the speeds-and-feeds of a given technology or system and very little on the use case. Is Fibre Channel faster than IP storage? Is IP storage easier to manage than FC? What about uptime and other high availability matters? And so on.
But if you look at the need for these technologies from the end-user perspective, the discussions become much more interesting. The new use cases for IP-storage, such as Big Data Analytics, Video Surveillance, Hyper-convergence, and Tele-Medicine applications, to name a few, have set the stage for renewed interest in converged storage networking.
With the introduction of the Cisco Cloud Scale (ASICs), Cisco is capable of delivering the next-gen architecture for any IP Storage network. These platforms provide the switching foundation to host the most demanding customer applications. Cisco Cloud Scale introduced a revolutionary technology called Smart Buffer in conjunction with QoS. Customers can now take advantage of these capabilities and build their IP/Ethernet-based storage networking based on their requirements with the Cisco Cloud Scale switching platforms.
Cisco advanced capabilities include:
- Smart Buffers for intelligent congestion control mechanism for efficient IP Storage
- Flexible performance options from 100 megabytes to 100 gigabytes in port speed
- Lossless fabric-wide architecture to support any storage traffic
- Seamless connectivity for distributed apps via container-based micro-services
Why Smart Buffering for scaling IP storage?
Let’s take a deeper dive into just one of these capabilities: smart buffering. When operating correctly buffering is essential to both maximize network capacity and minimize application and storage traffic latency ensuring the optimal application and storage response times.
With a Cisco Cloud Scale smart buffer approach, storage and network admins can manage buffers and queue scheduling intelligently by identifying and treating large and small flows differently. Smart Buffers provide intelligent congestion control mechanism and provide efficient IP Storage traffic management. It is not about having huge buffers, but it is about managing traffic efficiently. This contrasts with the “deep buffer” approach (most merchant silicon-based switches) all data is placed in a common buffer on a first-come, first-served basis, and then moved out in the same order – thus not enabling the network to distinguish between flow sizes. No matter how large the buffer is, the result of deep buffers is added latency for both large and small flows, benefiting neither.
And what about added cost? There is no question that adding more buffers will add unnecessary cost. It is estimated that switches with deep buffers can be three times more expensive, with no real traffic application performance advantage and in some cases detrimental by introducing more latency by starving certain applications. Cisco smart buffer approach allows large and small flows to share the switch buffers in a much more fair and efficient manner. This provides buffer space for small flows to burst and large flows to fully utilize the link capacity with much lower latency times – thus reducing the total cost of the solution.
Miercom tested this in its report “Network Switch Impact on Big Data Hadoop Cluster Data Processing”, which concluded that the network devices with adequate buffers and that are combined with intelligent buffer management algorithms handle traffic management proactively and efficiently with better application response time. This matters in terms of increasing not only the overall performance, but also in terms of extending the industry-leading availability and scale for enterprise storage traffic.
Even Dell/EMC’s recent design document, ScaleIO IP Fabric Best Practice and Deployment Guide – Dell EMC Networking Leaf-Spine Architecture for ScaleIO,” agrees with this approach. It details network switches in the test, (Z9100-ON with 16MB buffer as the spine and S4048-ON with 12MB) as leaf, that are not designed with deep buffers. Although they may have chosen other switches for testing, fact they did not consider deep buffers for ScaleIO is worth noting and an oversight for the needs of today’s data-heavy applications. By introducing deep buffers into these scenarios, would have negative impact on the application performance
However, by introducing Cisco Nexus Cloud Scale switches with Smart buffers, the total application response time can be reduced, thus increasing not only the overall performance, but also industry-leading availability and scale, for storage traffic. The Cisco solution not only provides smart buffers but nearly 100% uptime and options to scale, so customers can securely access data 24X7.
In general, the Cloud Scale numbers added up to be 50 percent lower in cost with 50 percent improvement in application flow time vs. comparable storage networking solutions. An agile storage infrastructure that is easy to manage, affordable and flexible is required to meet the needs of resource-constrained IT storage professionals. Cisco provides flexibility to design and build LAN and IP storage networking architecture for small and large customers.
Reference: Miercom | Network Switch Impact on “Big Data” Hadoop – Cluster Data Processing
Sr Marketing Manager
Great post Tony. Readers may also be interested in the excellent videos Tom Edsall created talking about Cloud Scale and how it delivers superior I/O performance for applications through dynamic packet prioritization and approximate fair drop in the videos found in this post https://blogs.cisco.com/datacenter/intelligentbuffer
Those two videos by Tom are great. Short, clearly articulated and based in fact. Worth a watch for anyone interested.
Comments are closed.