Gnodal from Arrow ECS

BUILT FOR SCALABILITY

Network growth is easily achieved by aggregating multiple GS-Series switches, supporting thousands of Ethernet ports while preserving Gnodal's superior performance characteristics. As systems grow, MAC address table size scales linearly with the number of switches, supporting up to approximately 52 million MAC addresses in total. A total of 16,000 multicast groups are also supported. Manageability of the whole fabric is provided by a single, redundant management instance. 

SYSTEM ARCHITECTURE

The GS-Series is the first offering in the Gnodal Data Center 10/40 GbE switch portfolio which is built upon the foundation of Gnodal's Peta ASIC, a fundamental differentiator from other switch providers. It is the Peta ASIC and its efficient integration within the GS-Series switches that provides a revolutionary Ethernet design that eliminates network congestion and delivers extreme performance at minimum cost and power consumption. As multiple switches are combined, the properties of low latency and line-rate throughput are preserved across multiple chassis, a clear distinction with Gnodal technology. Management of the switch infrastructure is also simplified through single-point network administration. This allows the GS-Series to extend the capabilities of the Ethernet standard into highly scalable configurations with minimal complexity.

ULTRA-LOW LATENCY

Small amounts of latency and jitter can have a considerable impact on application performance and ultimately quality of service delivered to the enterprise. Latency severely limits the overall scalability of applications and utilising the lowest latency is of utmost importance for customers wishing to guarantee and deliver optimal performance in their operations. Satisfying this requirement in larger configurations required considerable innovation. On ingress into GS-Series switching, the initial latency is sub 150ns with each subsequent Gnodal switch added to the fabric incurring only 66ns of additional latency. This results in an overall latency as low as 414ns for systems supporting thousands of ports with a rich fabric.

GUARANTEED MAXIMUM LATENCY

In many applications, providing ultra-low latency is only part of the answer. For example, in the competitive financial services sector, the lowest latency and guaranteed maximum latency with deterministic jitter is required for electronic transactions such as high-frequency trading in the derivatives and commodities markets. Gnodal offers organisations a strong competitive advantage by guaranteeing a maximum latency under load or by failures within the network. Variability in latency can also play a significant part at scale in media streaming or scientific simulation, typically manifesting itself in multicast or collectives traffic in which data is sent in a single transmission to many receivers and as such, is bound by the slowest member. GS-Series switches build upon the ultra-low latency features to encapsulate a highly efficient, non-replicating, multicast mechanism with minimal jitter.

LINE-RATE BANDWIDTH

The demand for bandwidth in support of I/O-intensive workloads is growing dramatically and is being driven by the adoption of server and storage virtualisation and the increased bandwidth demands of multi-core servers. For these requirements, 10 Gigabit Ethernet has become a natural choice and it has been extensively implemented in data centers, virtualised environments, internet content provisioning and co-locations. The demand for 40/100 Gigabit throughput has started and Gnodal offers full line-rate bandwidth across all ports within the GS-Series and provides arbitrary wide LAGs across switch chassis to provide extremely high bandwidth point-to-point connections, exceeding 1000's of Gbps channels between sub-domains of an enterprise.

MAXIMUM THROUGHPUT

For many applications such as in data-driven science/engineering as well as media and entertainment, organisations require a high throughput data network in order to deliver extreme and fair performance. Gnodal's solution offers organisations significant I/O capabilities in a converged network framework delivering high throughput under intensive traffic loads by alleviating congestion internal to the network itself, even in the presence of end-point congestion. Fair arbitration techniques further ensure that performance is balanced and hot-spots in the network are eliminated.

DYNAMIC MULTI-PATHING WITHOUT STP

One common criticism of Ethernet is that its Spanning Tree Protocol resolves loops at Layer 2 at the cost of performance as the network cannot fully utilise all available paths. Moving to Layer 3 switching improves scalability but also adds significant switch latency and load-balancing cost. GS-Series switches operate at Layer 2 and use a series of in-line and adaptive routing techniques not governed by off-loaded CPU cycles that avoid loops and therefore ensure high performance and full bandwidth utilisation. In addition to this, the GS-Series support L2+ enhancements such as IGMP and MLD snooping.

SOLVING NETWORK CONGESTION

How much of the system resources are actually available in normal day-to-day use depends on how the system behaves in the event of congestion. End-point congestion can be caused by network traffic, incast issues, or by faulty or broken hardware. In theory, these occurrences should not affect the rest of the network but in reality, congestion spreads quickly through the system creating hot-spot contention and the observable effects of higher latency, broader jitter, poor utilization and loss-of-service. Gnodal's Layer 2 solution uses a series of loop-free routing techniques that combine to provide distributed load-balancing across multi-chassis configurations, substantially alleviating network congestion caused by Head-of-Line blocking while maintaining strict in-order packet delivery. This distributed arbitration operates truly orthogonally to data transmission and avoids expensive and non-scalable off-load to external processing.

NETWORK CONVERGENCE

The GS-Series supports DCB/DCE providing per-priority pause in a lossless Ethernet capability with additional end-point notification. These mechanisms enable full support for RDMA and compatibility with other RDMA networks and Ethernet RDMA-capable NIC vendors. GS-Series switches enable support for FCoE and iSCSI making them an ideal choice for environments where IPC and I/O can be combined across a single converged network.

LOW POWER AND CONSOLIDATION

The high degree of integration in the Peta ASIC minimizes the number of components required within the network, resulting in a substantially reduced total cost of ownership, a competitive price per port, and exceptional power efficiency. The GS-Series also provides a cost-effective migration path from 10 to 40 GbE that protects existing network infrastructure investment. As the enterprise migrates to 10 GbE client access points, consolidation and storage deployments naturally migrate to higher speed aggregation. The GS-Series provide this capability implicitly, providing organisations the necessary flexibility required for consolidation.

STANDARDISATION

The GS-Series employ industry standard SFP+ and QSFP connections, supporting active or passive copper twinax cables and SR, ER and LR optics for a range of deployments. Interoperability is ensured by providing standard L2 Ethernet protocols at the edge. Switch management is provided by the Aricent™ Intelligent Switch Solution (ISS)™ which provides an extensible and reliable protocol stack from an industry leader in this field. Gnodal ISS has extended the capabilities of this stack to provide a single logical switch management infrastructure spanning multiple switch chassis. This vastly improves administration of complex switch deployments. Extensive, standards-based protocol support is provided within a familiar CLI for management and security together with a flexible API for custom integration.

DEPLOYMENT EXAMPLE

The GS-Series with its combination of 10 and 40 GbE provide incredible flexibility, allowing network managers to achieve the correct balance of port count, performance and price. A myriad of efficient topologies to suit all applications are possible and range from:

• Simple star-based topologies

• Grids and hypercubes maximising connectivity

• Fully connected fat-tree networks maximising bi-sectional bandwidth and high availability

GS-Series Example with multiple edge GS4008 switches connecting servers and storage across a GS0018 fat-tree switch fabric.

Larger networks can be constructed by adding additional stages. The GS-Series extends low latency across multiple stages with 78ns per stage and preserves high utilisation across multiple switch chassis. 

SUMMARY