December 02, 2010
The folks at CERN are upgrading their high performance network infrastructure and for the first time are bringing Brocade Communications into the mix in a big way. CERN's IT group is set to deploy Brocade's new MLXe Core Routers in order to support the voluminous and growing storage and compute capabilities needed to keep the organization's datacenters humming.
In case you've been on another planet for the past few years, CERN, aka the European Organization for Nuclear Research, has been busy trying to uncover the mysteries of particle physics in order to discover some of the fundamental laws of the universe. Using its Large Hadron Collider, the world's largest high-energy particle accelerator, physicists have been conducting experiments that are generating scads of data -- on the order of 15 petabytes per year -- which CERN is streaming to its own Tier 0 datacenters in Switzerland and France, and then on to Tier 1 partner centers around the world.
Although CERN drives the global LHC computing grid, it alone is responsible for the five Tier 0 datacenters, which get first crack at the raw data, and it is here where the Brocade routers will be deployed. Four of these are smaller facilities dedicated to performing the initial number crunching on the collider data. After this first stage of processing, the results are sent to the main CERN datacenter for storage and reprocessing, and subsequent distribution to Tier 1 partner sites. In the main center itself, there has 8,000 servers that encapsulate 40,000 cores, as well as 40 petabytes of disk and 45 petabytes of tape.
As you might imagine, processing and shuffling all this data around requires some serious networking prowess. At CERN's Tier 0 centers, the current non-blocking capacity of the network core is around 4.2 terabits per second (Tbps), and is projected to hit more than 10 Tbps in 2011. "We are basically doubling the capacity of the system roughly every two years, " explained Jean-Michel Jouanigot, communication systems group leader for CERN's IT Department.
To accommodate this kind of growth curve, CERN realized they would need core routers that would be 100GbE-capable. Jouanigot told HPCwire that the selection of a vendor that could deliver this kind of capability began last year. The process was fairly straightforward: CERN had an extensive list of requirements, the primary one being the ability to get to 10 Tbps with a year. After in-house testing of manufacturers' gear that could meet these demands, it came down to price. As Jouanigot put it, the Brocade proposal was "the least expensive technically acceptable solution." The company's MLXe gear will eventually displace the Force 10 routers that currently predominate at CERN.
Although the goal is 100GbE networking, the Brocade routers delivered to date sport 10GbE ports. The MLXe routers support the higher data rates, but Jouanigot is still waiting on the company to get the pricing info on the 100GbE version, which are not yet available in quantity.
He says the problem they currently have with their 10GbE network core is that each router is connected to every other router through 16 10GbE ports. This entails a lot of cabling and additional management, so going to 100GbE to simplify the networking setup just makes sense. But according to Jouanigot, it really depends on how Brocade decides to price the 100GbE ports; if it turns out to be more than 10 time the 10GbE price, that becomes a tougher decision, extra cabling and management notwithstanding.
Another challenge is that when they renew their compute servers, they need to decide if they want to move to 10GbE system interconnects. Today, CERN servers are predominately GbE connected, with just a few hundred 10GbE-attached. It is very likely that newer servers will be delivered with 10GbE NICs, Since these systems are hooked up to one another, this will further draw on the bandwidth of the core network.
The routers also drive the external links between the various datacenters, so going to 100GbE will allow them to accomplish faster inter-center networking without having to lay new optical fiber. That tends to be a very costly operation -- yet another motivator to up to 100GbE. Jouanigot is optimistic that he'll get his faster ports, though. "For us, I see the introduction of 100GbE at CERN very likely next year."
Posted by Michael Feldman - December 02, 2010 @ 4:40 PM, Pacific Standard Time
Michael Feldman is the editor of HPCwire.
No Recent Blog Comments
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?