Interconnect at speeds reaching 100Gb/s, are taking the data center network world to never seen before scale levels. The requirement from Web 2.0 market drivers call for the introduction of new technologies to allow for the network to function at these scales. While some devices claim support for such protocols as VXLAN or MPLS, it is the support of these protocols at scale that actually matter.
Mellanox’s introduction of the new switch ASIC named Spectrum is doing exactly that. With extended table sizes, advanced hashing mechanism for high utilization rates of the table as a resource and optimized balancing capabilities that avoid duplications of table entrees, it allows for an MPLS network to grow to levels as required by the giants of the industry.
Together with lowest latency, predictable cut through switching, and the lowest power consumption in the market, the new ASIC also brings back some old news of meeting the benchmark performance test known as RFC2544. Fulfilling the world’s de-facto performance benchmark indicator becomes news again and is the subject for the rest of this blog.
PACKET LOSS IS UNACCEPTABLE BEHAVIOR
While it is obvious that losing customer data is a taboo in any provider-consumer relationship, it has become a fact that in many data center switch designs, packets are routinely being dropped in non-congested scenarios. While creating artificial blocking ratio on the top of rack level is anticipated and controlled by the user, switches secretly lose packets due to design compromises made by ASIC vendors and kept hidden behind flashy software goodies. It is creating a problem for data using applications and impose high costs via software complexity, increased and unpredictable latency and CPU consumption–all leading to application performance degradation.
The fact that this design flaw is kept hidden also means the loss of the packet is totally random and cannot guarantee priority packets being served accordingly. The impact on network behavior can range anywhere between negligible loss of best effort traffic to paralyzing critical applications and no ability to predict or guarantee a service level agreement from the data center. However, as said by the great playwright, William Shakespeare: “It is not in the stars to hold our destiny but in ourselves.”
The Mellanox Ethernet switch portfolio, based on Mellanox own ASICs–SwitchX, SwitchX-2 and Spectrum–provide line rate performance with even the smallest packet size, at any port count and with any traffic pattern.
As trivial as it might sound, a system not losing packets has a direct impact on application behavior. The result is database, big data analytics and cloud workloads that run more efficiently and reliably. This high level of predictability also enables your workloads to scale out as your business grows and prevents you from making unnecessary compromises.
Zero packet loss also means non-compromising cut through technology independent from other workloads or ports speeds used with the system. While other systems silently degrade to store and forward behavior on the complete system level in cases where different port speeds are being used, Mellanox switches independently optimize each traffic flow to best cut through behavior that provide predictable latency and as a result, optimized application performance.
Looking at increasing speed of storage technologies, latency on the switch level becomes a real bottleneck if slow and jittery. On the other hand, it becomes a real technology enabler if latency is low and static. Mellanox switches are the only Ethernet switches in the world that can provide a network solution that can grow from disk to flash and onto benefiting from the ultra-low latency and high transaction rate of advanced post flash storage technology.
An additional benefit of Mellanox Ethernet is the support for 56GbE interface speed. This non-standard Ethernet speed can only operate between Mellanox devices but together with the complete end to end Ethernet solution from Mellanox, provides a bandwidth boost of 40% on top of the standard 40GbE line rate which is common to the QSFP interfaces.
This extra bandwidth can allow for higher application performance as well as absorb various hiccups while keeping applications tuned to the standard 40GbE line rates as well as allowing the widest bandwidth pipes between current generation interfaces to future generation 100GbE systems.
The obvious question here would be “Do I need to pay for this premium service and how much?” and the answer is ‘No’. With Mellanox Ethernet switches, you get more network and application performance value while paying less for networking gear.
Be the master of your own destiny and enjoy non-compromising Ethernet behavior with Mellanox switches. (Tweet this!)
You can start benefiting from the most efficient data center today. To learn more, read the Tolly report on mellanox.com.