- Category: Features
- Published on Thursday, 13 December 2012 11:25
10 Gigabit Ethernet is steadily becoming the norm in enterprise networks. Will 40G and 100G follow suit or take time to catch on? by Harshal Kallyanpur
Cloud computing, virtualization, enterprise mobility and Big Data hog the limelight in any conversation about enterprise IT. Finding little mention is the network, which has been around for the longest. Slow and steady on the growth track, networks have seen innovation, which is staggered in comparison to some of these technologies.So, where does it stand today? Connectivity at the consumer side has gone way past the 56k modem phase and consumers today are experiencing broadband speeds that go up to a few Mbps easily. The story on the enterprise side is also similar. What used to be Mbps yesterday is Gbps today. Organisations which, till a while ago had networks which had fast Ethernet or one Gigabit Ethernet connectivity are today looking at 10 Gigabit (10G) Ethernet as a standard for connectivity within the enterprise. Some are even talking about 40G and 100G at the data centre side.
However, network infrastructure in any organisation is deployed in a way that it sits in the organisation for the long haul. Putting a network in place, means pulling a lot of cables across and through walls, through floors, around furniture and even across buildings. And then there are switches and boxes which need to be intelligently connected to these wires. Therefore, the network, is usually preferred to be an effort that does not require frequent revisits.
Applications call for faster networks
A few years ago, a typical IT infrastructure meant a few servers hosting basic applications such as e-mail, finance, and ERP. These applications were spread across servers that made up for a data centre environment in a small room. The workloads that these applications generated were nothing much to write home about, when compared to the current generation applications, and hence could run easily with the then available compute, storage and network resources.
However as businesses started growing, enterprises started to increasingly leverage on IT to drive their business. Applications grew complex, and automated a lot more business functions.A bigger and more complex application meant more workload. Most enterprises then were running networks that offered speeds of a few Mbps at the user access level and one Gigabit (1G) Ethernet speeds at the distribution and core level. Surging application requirements compelled them to look at going one level up on the access side, the distribution and the core.
Says Prem Nithin, Business Development Manager, Borderless Networks, Cisco India, Cost enterprises were using 10/100 Mbps links at the access side and application requirements were not very high. However, Fast Ethernet started becoming a bottleneck as applications grew. Furthermore, most organisations traditionally had applications running on a distributed architecture. This model was not scalable and organisations started thinking about consolidating applications. One Gigabit was proving to be insufficient and therefore 10G started finding relevance at the server level.”
Sharing a similar opinion Bhavin Barbhaya, Vice President - Sales (Network Infrastructure), AGC Networks says, The convergence of voice and data as network traffic on the same network created the need for going from 100 Mbps to 1G and from 1G to 10G Ethernet. Most large organisations today have moved to 1G at the access level and 10G at backbone level.”
10G Ethernet finds more takers
Applications over the last few years have increasingly been going web-based or online. They have also transformed from being business-enablers, to being business-drivers. E-mail today integrates with instant messaging, ERP has grown beyond finance, HR, SCM, to functions like CRM and Business Intelligence (BI) and servers are no longer boxes of applications running in silos.
Says Prakash Krishnamoorthy, Country Manager, HP Networking, HP India, ?bout 10 years ago, we saw a rapid proliferation of desktops and these desktops needed Ethernet connectivity. Also, organisations adopted a lot of servers and typically hosted applications such as ERP, mail messaging and file services across these servers and they were discrete and few.”
“Today they have grown from ERP to CRM and data warehousing, and gone from single instance mail messaging to multi-instance messaging. There is an increase in the number of servers today, which in turn has forced the need for virtualization.”
All this created the data traffic within the enterprise, so large that the then current generation networks could not handle it. Enterprises are today moving to 10 Gigabit Ethernet, especially on the network and the server side, where the need for faster connectivity is felt the most. And One Gigabit Ethernet is becoming standard at the user access side.
Vendors believe that many organisations that had invested in network technologies to last them for four to five years, are today looking at refreshing their network infrastructure. And when doing so, they are ensuring that they at least have 10G Ethernet connectivity on the network and the core level and are trying to put as many servers on 10G Ethernet as possible.
On the other hand, all greenfield projects will look at 10G Ethernet connectivity, at the distribution and server level ground up, while making them future proof and ready for the perceived 40G Ethernet requirements coming in.
According to Subhashini Ramakrishnan, Chief Technology Manager, Dax Networks, 10G Ethernet is becoming the de-facto standard for server-storage communication. With servers and storage growing in capacities, performance and workload, there is no point in having huge capacities if the access pipes are not as large.
She further observes that 10G Ethernet connectivity is becoming important even on the network side and cites example of the healthcare industry where areas such as Imaging, Cat-Scan etc. are generating huge amount of data that is flowing over the network. Data of this kind, typically runs into Gigabytes and networks with 10G Ethernet connectivity would be able to send across these files in a matter of few seconds.
However, she is quick to point out that cost has also been a key driver for the change in the level of adoption of 10G Ethernet. Subhashini says, ? lot of India is still on 100 Mbps. Large scale adoption of 10G has only happened in the last two years. If the cost of 10G comes down by 25 to 50%, the technology will see further adoption.”
Giving his perspective on this, Nithin of Cisco says, G Ethernet for today is more or less sufficient at the access level. 10G would be an overkill at the access level at this point. It is not a mass market requirement from an access point of view, with very few industry sectors requiring it. And while the cost of 10G connectivity has come down, it has not come down to a point wherein it can be provided at an access level.”
Sharing a similar opinion, Jeevan Belliappa- Global Business Manager ITS, Wipro Infotech says, Enterprises were talking about moving to 10G almost three to four years ago, but the pricing was exorbitant. Same was the case with 1G Ethernet around ten years ago. However today most laptops and desktops come with 10G ports. It will take another two years for 10G Ethernet to peak, and once that happens it will open the doors for adoption of 40G Ethernet.”
Setting the stage for 40G Ethernet
While most of enterprises are still grappling with 10G Ethernet network deployments, 40G Ethernet is already cropping up its head, or at least the vendors believe so.
A major reason for this is being attributed to the fact, that many organisations are consolidating their vast, underutilized, over-provisioned server sprawls. And these consolidating server farms are seeing inter-server data communication, that the current 10G Ethernet interconnects may not be able to handle.
Furthermore, the consolidation on the server side has also resulted in centralisation of storage. The same 10G Ethernet network is now expected to interface with storage that hosts several hundred servers as virtual machine files and their associated data.
Server racks today have every physical server with a dedicated connectivity to a Top-of-the-Rack(TOR) Switch. The moment the server hardware is virtualized the workload of say for instance, 100 virtual servers would be running across 20 physical servers. These 20 physical servers are now expected to have a bandwidth for 100 servers.
Currently many organisations are running servers with 10G Ethernet connectivity between the physical servers and the TOR switches. However, the amount of data generated by the virtual machines, the inter-server communication between them and the virtual machines themselves as a file, are putting tremendous pressure on the interconnect pipes, and thus, creating a bottleneck at the network level.
Says Ashutosh Kapoor, General Manager - Head Alliances, ITS, Wipro Infotech, ?ervers have become much more intelligent, and divide resources between multiple virtual machines running several different applications. The bottleneck would be the network and latency earlier was much higher. The need for high-performance networks was felt as data centres started becoming more intelligent.”
He further explains and says, While the latency might not have changed between then and now, the number of applications that are being accessed or the amount of workload running on through these pipes and intelligent servers and storage boxes is a lot more.”
As an answer to this, most organisations are aggregating several 10G Ethernet links together, to achieve a combined higher throughput, with redundancy and load balancing at the server level. However, in the long run, at some point they would need hardware which can give a single throughput of 40G or more.
According to Nithin of Cisco the entire journey from consolidation, to virtualization, Cloud computing private, public or hybrid, has taken two to three years. Enterprises then were only talking about it, and today they are increasingly adopting these approaches. These workloads are now demanding more throughput at the data centre level and it is now that enterprises are looking at if 10G would be enough or whether they would need to upgrade to 40G.
He is quick to add that access to distribution is still a Gigabit link, while distribution to core is 10G. It is only when the upgrade to 40G happens at the data centre that is when organisations will start looking 10G for distribution to access.
Mirroring this opinion, Krishnamoorthy of HP says, G and 10G are perceived as the right technology for client-server connection. Adoption of 40G at the rack is seen as a visible trend. Most servers come with a 10G port and multiple such ports terminate into a highly density TOR switch. The connectivity between TOR switches on multiple racks or connectivity between the TOR switches and the core network is happening using 40G.”
The roadmap for 40G, and 100G Ethernet
Though 40G Ethernet seems to be just around the corner, it will take some time before, 40G sees good levels of adoption. There are some compelling reasons that will limit adoption for the next couple of years.
For instance, although the standards are in place, and most vendors are supporting 40G and even 100G in some cases, there is no large scale requirement for the technology. Most of the requirement is coming at the data centre and core connectivity level.
Kapoor of Wipro says, Telepresence and such other applications are yet to penetrate enterprises. However, when they do, there will be a huge demand for network connectivity. It is a chicken and egg situation wherein, the enterprise upgrades the network to meet the demand, or meets demand because the network is equipped to take it on.”
According to him, currently media houses have shown interest for 40G and 100G as today these organisations are streaming programs live on web, especially for audiences abroad. They also have requirements for live streaming which is periodic, such as cricket matches. These requirements are driving interest in 40G and 100G.
He believes that vendors are equipped to address the requirement, through basic software upgrades for their devices that can enable network devices such as switches and routers to use the same network to deliver faster speeds.
Nithin, however, believes that while the standards are in place, and devices support 40G, a throughput of 40G per port in reality, is still debatable. Vendors are currently bundling multiple 10G links to deliver a 40G throughput.
Enterprises too are asking for support for 40G and showing interest in 100G Ethernet at a concept level. However, they are not investing actively in these solutions, but choosing to take multiples of 10G connectivity to arrive at 40G or higher.
The reason for this is the fact that at this point, 40G would be an expensive proposition and only a few vendors would be offering specific solutions. Furthermore, there isn? enough requirement within an enterprise that the IT department would able to justify the cost of deployment of 40G across the organisation, to the management. Therefore, the market for 40G is currently not big enough for vendors to offer products that give true 40G Ethernet or higher speeds per port and hope for organisations to adopt the same.
Subhashini of Dax says, It is a typical mentality among Indian organisations to not pay for something they will not use immediately. Therefore, the actual shift to these technologies will only happen when applications demand this kind of connectivity.”
Giving his perspective on cost, Barbhaya of AGC Networks says, Two years ago 10G Ethernet was expensive. It is a volume game and early adopters (of 40G Ethernet) will end up paying a higher price. However, whether it is currently required or not, is something that organisations should ask themselves. If it is a strong requirement, cost should not be a major concern.”
According to him, organisations doing a hardware refresh or buying hardware for a greenfield project, are looking at whether the equipment that they are buying, are giving a roadmap for 40G or 100G and if they support these speeds. Two years down the line, moving to 40G should simply mean changing a module and not the entire infrastructure.
HP on the other hand believes that the move from 10G to 40G will happen faster than the move from 1G to 10G, particularly at the data centre and the core level. Krishnamoorthy says, Earlier customers would buy TOR switches with 1G connectivity with 10G uplink. Today 50 % of the customers are asking for TOR switches with 10G connectivity and 40G uplink. The availability of ports and transceivers on the server side has also changed dramatically.”
40G Ethernet, therefore will need to wait for a while until it can go mainstream. It will take a while before data centre requirements propel the need for 40G, and it sees enough large volume adoption, such that prices in turn come down. 100G Ethernet is still at a concept level and a little far off from seeing any upfront adoption.
Currently, 1G Ethernet is serving the requirements of the enterprise at the end user level quite adequately, while 10G Ethernet is holding up strong at the network and server level. It will be slow and steady for next few years, until each of them find themselves being phased out by the other higher bandwidth standards.