Broadcom’s Smart Buffer Technology can provide cost effective performance scaling of cloud applications in Broadcom StrataXGS data center switches.
Designing systems for Cloud Data environments that perform well and are cost effective is essential to assuring that the system integration with Broadcom SDKs work seamlessly in a cloud environment.
The function of a network switch is to receive packets on an input port, apply switching and routing decisions as per the configurations that are set on the L2/L3 switches, identify the outgoing port(s) and send the packet out.
Whenever a switch receives a traffic burst, the output port will not be able to process the complete traffic as it is receiving on the input port. As a result, we need to implement store and forward logic at the input port to have lossless transmission of the packets.
The buffering capability of the input port determines how many packets can be buffered by the port before the output port could send the packet out of the switch. Whenever the burst is high/continuous, it is very likely the buffering capability of the input port could be exhausted and packets could get dropped randomly. This results in poor and unpredictable behavior.
The buffering capability of a port is determined not only by the amount of buffer space allocated for the port, but also by the Memory Management Unit architecture which plays a vital role in absorbing such types of burst in real time.
In cloud-based data centers, it is evident that “Bursty” traffic patterns are very high. Therefore, it is key for anyone who is designing a cloud-based data center to have good buffering algorithms implemented in the switches, so that it could absorb such traffic to its maximum capability. This approach could minimize the loss of traffic and as a result have better connectivity and reliability.
The traffic patterns in a cloud environment are typically dynamic and uncertain. Examples of this type of traffic are Hadoop MapReduce, distributed file systems used in Big Data analytics, distributed caching related to high performance transaction processing, streaming media services, and many other demanding and high bandwidth computing processes.
It’s better to take a close look at the traffic pattern characteristics in the context of Big Data such as Hadoop and MapReduce since these comprise a high percentage of the traffic in the data centers.
In Hadoop File System (HDFS) operations such as input file loading and result file writing, give rise to network burstiness due to a high amount of data replication across cluster nodes in a very short time span.
The data shuffling phase in MapReduce also tends to create many-to-one bursts when multiple mapper nodes terminate and send their results to reducer nodes in the network.
Even though it is recommended to migrate from TCP collapse to Priority Flow Control (PFC), Quantized Congestion Notification (QCN), and Data Center TCP (DCTCP), these protocols are not widely implemented in today’s Web and other cloud-based networks. That’s because it would require a costly and complex upgrade of the hardware and software in multiple nodes of the network. As these protocols cannot solve the complete problem, they handle short lived traffic flows or micro bursts.
One solution for the above problem is to use an underlying platform supported framework like Broadcom’s Smart-Buffer technology. It offers a proven approach for delivering cost-effective packet buffer performance and is well suited for modern data center switches running cloud applications.
The StrataXGS switch architecture with Smart-Buffer technology incorporates a scalable multi-pipeline design interconnected through a centralized MMU architecture. Further, its packet buffer is right-sized and dynamically shared across all ports for excellent burst absorption.
Its architecture enables global admission control, queuing, policing and shaping functions. Smart-Buffer delivers optimal buffer utilization and burst absorption for data center workloads by taking a holistic approach to buffer management – using real-life data center traffic scenarios to maximize overall throughput and lossless behavior. To summarize:
- Excellent burst absorption
- Fair shared buffer pool access
- Port throughput isolation
- Traffic independent performance