The Portland Group
Oakridge Top Right
HPCwire

Since 1986 - Covering the Fastest Computers
in the World and the People Who Run Them

Language Flags

Visit additional Tabor Communication Publications

Enterprise Tech
Datanami
HPCwire Japan

Arastra Starts with a Name Changer, Aims for a Game Changer


Arastra, a startup that introduced its high-performance 10 Gigabit Ethernet (GbE) switches in 2007, has changed its name to Arista Networks, and with it, comes a new focus on the emerging cloud computing market. The company has tapped former Cisco VP Jayshree Ullal as the president and CEO, and Arista co-founder Andreas (Andy) Bechtolsheim as the chief development officer and chairman. Bechtolsheim, who also co-founded Sun Microsystems, will give up his position there as chief architect, but remain on as a part-time advisor.

Bechtolsheim, along with chief scientist David Cheriton, co-founded Arista and are providing all funding necessary to get the company off the ground. Through the years, both men have accumulated considerable wealth through shrewd investments in technology startups, and are each thought to be billionaires today.

Bechtolsheim and Cheriton were two of Google's original investors, each reportedly investing $100,000 in the search company startup in the late 1990s. They're not revealing how much has been invested in Arista, but their commitment is to fund the company until it goes public. "We're re-investing some of our Google proceeds in what we think is a very promising new opportunity," Bechtolsheim told me.

Their competition includes Cisco, with its new Nexus 5000 10 GbE switches, as well as similar gear from Force 10, Extreme Networks, Fijitsu and Blade Network Technologies. In general, these are more full-featured boxes, offering support for additional protocols such as FCoE (Fibre Channel over Ethernet) and DCE (Data Center Ethernet). The added functionality tends to make these general-purpose switches more expensive, and superfluous for Arista's target domain: ultra-scale datacenter applications.

A couple of weeks ago, Woven Systems introduced the TRX 200, a 24-port 10 GbE switch that matches up pretty well with Arista's 7124S switch. And unlike the rest of the competition, the cost of Woven's switches are more in line with the $400 per port pricing Arista is quoting for its 7100 series.

Where Arista heads into new territory is with its Extensible Operating System (EOS), a special layer of software that rides on top of standard Linux Fedora in each box. It allows third party software to mingle with EOS, giving users an opportunity to add new protocols, customized network management, or even more exotic applications to their network. It will also makes it easier for Arista to add its own features to EOS in the future, since it doesn't have to deal with a proprietary OS-kernel or software components that it can't control.

EOS also has a self-healing feature due to its modular design and protection mechanisms. All of the software processes are put in their own protected address spaces, and if one of those processes fails, it will automatically restart without dropping packets. This kind of robustness also allows for on-the-fly bug patching and software upgrades. Some high-end Cisco boxes have similar functionality, but in a much higher price band.

Arista started shipping gear in May and has enjoyed some success across a range of markets including financial services, manufacturing, government, education and healthcare. Some of its current customers are BitGravity (video content distribution), Northwestern University (VLAN services), and Lawrence Livermore National Laboratory (petascale computing, but probably 10 GbE just on the storage side).

According to Bechtolsheim, the level of reliability at this price point is unprecedented in the industry. "It was architected and implemented in a way that makes it more robust than any product out there -- and I'm not making that statement lightly," he says. According to him, the combination of price, performance and extensibility makes it a natural for cloud computing and HPC. In the latter market, Bechtolsheim believe most of the 10 GbE opportunity is on the storage side, where in many cases, storage servers are hanging off an Ethernet LAN and still need to talk TCP. On the compute side of HPC, he thinks InfiniBand will continue to maintain its position, due to its superior performance for MPI-based workloads.

But the company sees a much bigger play in cloud computing, where 24/7 mission-critical service using cheap, fault-tolerant hardware is the goal. Today, with the ubiquity of multimedia and the increasing density of compute power in servers, 10 gigabit pipes are now needed to keep up with all the data traffic. But since most of the software is running on top of a TCP stack, 10 GbE, not InfiniBand, is the obvious answer.

Right now there's a lot of activity from 10 GbE vendors, as the technology starts to move into the datacenter. Most analysts think that the transition from GbE to 10 GbE is going to occur fairly quickly over the next couple of years. Today the average selling price for a Gigabit Ethernet port is around $150. According to Mark Foss, Arista's marketing director, once the cost of 10 GbE approaches a reasonable premium over Gigabit Ethernet, the market will start transitioning very rapidly.

In the next year, as 10 GbE interfaces become standard on server motherboards and the next generation of CPUs get deployed, motivation to upgrade to the new standard will be much greater. As this occurs, more Web-based software services will start appearing to take advantage of 10 GbE performance. "There are a lot of things happening in the market -- in addition to the price points -- that will make 10 Gig to the server a no-brainer in the very short term," says Foss.

Most Read Features

Most Read Around the Web

Most Read This Just In

Most Read Blogs


Sponsored Whitepapers

Breaking I/O Bottlenecks

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.

A New Ultra-Dense Hyper-Scale x86 Server Design

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.

Sponsored Multimedia

Xyratex, presents ClusterStor at the Vendor Showdown at ISC13

Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.

HPCwire Live! Atlanta's Big Data Kick Off Week Meets HPC

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?

Xyratex

HPC Job Bank


Featured Events


HPCwire Events