2013 Predictions for Networks
December 28th, 2012 By: Industry Perspectives
Atchison Frazer is the CMO of high-speed networking gear provider Gnodal.ATCHISON FRAZER
2013 brings with it more innovations in the networking world. As demands for faster and bigger connections continue apace, the technology will be evolving to meet that demand. Here’s a few thoughts on what we expect to see in the New Year.
Big Data = Fast Data
Big Data analytics performance requirements challenge conventional data center network architecture (three-tier, oversubscribed, multi-hop schemas unnecessarily introduces server-server latency). A majority (better than 55% chance) of HPC-like environments will adopt Ethernet fabric architectures at both the network and infrastructure levels to meet the performance demands of enterprise Big Data. These requirements include line-rate, non-blocking and any-to-any communications between devices with massive amounts of flat/leaf-spine East-West traffic optimized for Layer 2.
Virtual Switches Enable Infrastructure Fabrics
Infrastructure Fabrics will be predicated on an increase of up to 10X of server and storage capacity, not just current requirements. They will need to scale beyond the standard VMx8 topology and maintain persistent interlink communications to establish a Fabric-of-Fabrics (FoFs) topology for über-management tracking of performance metrics and scalability risk factors.
Ethernet-as-Transport (EaT) will, by necessity, become a reality. Conventional, often proprietary, data center bridging techniques that are striving to scale Petabytes of capacity, suffer performance and congestion issues due to sub-optimal path-selection algorithms (static hash-based vs. adaptive real-time); slow and unreliable convergence mechanisms upon link failure (lack of cohesive fabrics); and under-utilized bandwidth (conventional switch capacity yields less than 30% utilization).
TRILL is Shrill
TRILL bridging will lose favor to All-you-can-EaT fabrics, given that TRILL requires too many changes to path selection and topology, relies upon static algorithms, and not least, lacks pre-emptive mechanisms for avoiding congestion. EaT Fabrics, by necessity ASIC-accelerated, will gain currency by delivering higher performance Ethernet communications with ultra-low latency and infinite scalability benefits. They are deterministic and offer dynamic multi-path bridging within the fabric that converges at an order of sub-100 nanosecond transmission speeds across the entire fabric.
Proprietary Ultimately Loses in an iOS/Android World
Networking standards will ultimately succumb to cultural and consumer technology habits, giving way to open standards and dictate convergence. For example, an emerging trend is to leverage the inherent extensibility of TCP to build first-class solutions that work in multi-tiered switching LAN environments without the need for legacy switches. Look for enabling protocols to gain currency, such asThe iWARP protocol, which is routable, robust against packet loss, and enables data flow over long distances, so it can be used in routed LAN/WAN environments and highly compute and data intensive applications that require bandwidth on demand.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.
All of what you are saying is very intriguing. I do think that ethernet transport (to customer) is the only way that will soon exist. SONET Fiber line cards remain expensive because of the math involved with BIP 1-3 error checking and intrinsic L2 routing. SONET’s lack of use outisde of meet-me rooms and carrier hotels will limit the ubiquity and thus the economics of scale as a interconnect. But you said TRILL as a whole is going away, I disagree if only because most manufacturers are heading away from spanning tree. TRILL (http://en.m.wikipedia.org/wiki/TRILL_(computing) ) is part of this predicate. Basically a bunch of bridging links (for various protocols and layer 3 routing) that have a deterministic engine for that fabric. I dont know if Gnodal switches use SPB or STP still but I do agree that some manufacturers may have shot themselves in the foot with customers with STP-less designs (I know a few Fortune 500 companies that have had their entire infrastructures come to a screaming halt with V1 series of certain chassis). But the TRILL and SPB ideas are sound (in general) especially with accelerated cache fabrics. Radia Perlman would even tell you that STP is troublesome in large topologies. A mixed L2-L3 route caching engine helps you scale the number of nodes and paths. Also, I would never have the guts to put iOS in the same sentence as “Open”.