Simon Bullen's Weblog

John Roese Nortel CTO

John Roese (Nortel CTO) has joined the 21st century and started a blog, so far the entries have been interesting and I think will grow more interesting over time. His last entry is regarding the amount of devices that are going to be on the WWW and how machine traffic will exceed human traffic, he call this phenomenon "Hyper Connectivity".
While I agree with his train of thought I must question the volumes he is anticipating.
We are seeing a trend where multiple devices such as phone, MP3 player , cameras etc are all becoming a single unit there fore only one address is required for multiple functions. Also the explosion in server virtualisation with Solaris containers and VMware is dramatically reducing the volume on machines in the data centre.

I am interested in Johns thoughts towards Eco friendly hardware and Nortels direction on the virtual server / switching space, maybe I need to drop him a mail............

I have added Johns Blog link to my book marks for those of you who want to read more.

Posted at 09:21AM Feb 06, 2007 by Simon Bullen in Networking  |  Comments[0]

Cisco enters digital signage market

Comms giant Cisco has moved into the digital signage market with the launch of a Digital Media Manager system and a Digital Media Player for businesses.

Cisco’s digital media to business vice-president, Thomas Wyatt, said, “This is a comprehensive system for playback of high definition media over IP networks on LCD and plasma displays. These products are enhancements of systems we acquired after our takeover of digital signage vendor Tivella in December 2006.”

Cisco has signed a deal with LCD vendor NEC Display Solutions to use its monitors, but will use its own reseller channel to market the digital signage products.

Wyatt said that with the cost of LCD and plasma technology falling, and investment by advertisers in in-store media growing, the value of the digital signage market is expected grow from about $400m today to an estimated $2bn by 2010.

“It’s best to get [your advertising message] to customers when they’re thinking of purchasing products, which means when they’re in the store. Basically, we’re getting the right info to the right customer at the right time,” said Wyatt.

The Cisco Digital Media Player is an embedded Linux device weighing about one pound and has 1GB of Flash storage, good enough for one hour of standard definition video or 30 minutes of high-definition video.

The Digital Media Manager (DMM) is software that can run on one of Cisco’s Media Convergence Servers, similar to that running Cisco’s IP Telephony software. DMM comes pre-installed, and can be used to manage thousands of media players. After the initial configuration files and environmental variables are set up, the system could be up and running within 30 minutes, Wyatt said.

Wyatt added that the lack of a hard drive made the Digital Media Player a more secure and reliable platform for digital signs than PC-based alternatives

A 50-media-player bundle with associated hardware and software costs around $150,000 (£76,000 + VAT), said Wyatt.

Posted at 10:32AM Jan 25, 2007 by Simon Bullen in Networking  |  Comments[0]

Cisco buys IronPort to push into anti-spam market

Networking giant Cisco Systems has agreed to buy email filtering and web security firm IronPort Systems for $830m in cash and shares. The deal, expected to close in Cisco's third quarter ending April 2007, pushes Cisco into the anti-spam market for the first time. The acquisition is Cisco's fifth largest to date, Reuters reports.

IronPort, founded in 2000 and with a workforce of around 408, markets a line of IronPort anti-spam and gateway security appliance appliances as well as running an email reputation service called SenderBase. It also sells technologies designed to limit the spread of viral outbreaks.

IronPort recorded sales of $25m for the quarter to August 2006, its last available figures, so Cisco's purchase price would appear quite high even though it has a long history of successfully using acquisitions to boost its profitability by entering adjacent markets in networking and security.

Although well-known for its PIX firewall, VPN, network access control, policy management and intrusion prevention products, Cisco has previously held back from entering the market for technologies that filter out junk mail.

Cisco said that it would retain IronPort's existing channel structure post-acquisition, when IronPort will become a business unit in Cisco's Security Technology Group. ®

http://www.theregister.co.uk/2007/01/04/cisco_ironport_purchase/

Posted at 03:04PM Jan 04, 2007 by Simon Bullen in Networking  |  Comments[0]

Infoworld technology awards - the other winners

While everyone is raving about Sun's awards for the X4200 and X4600 m2 under the hardware category, I decided to take a look at what won in the Networking section.

Riverbed Steelhead 3.0 won the best WAN accelerator:-
Perennial winner Riverbed continues to impress with expanded CIFS and NFS support in 3.0, allowing for better performance with both Windows- and NFS-based servers. QoS can now be enforced in the appliance with policies based on bandwidth and latency. Detection and correction of asymmetric routes is also a big plus.

Extreme BlackDiamond 10808 and 8810 won the best 10Gb switching infrastructure:-
Extreme BlackDiamond represents the Cadillac of Ethernet switching infrastructure. The 10808 core chassis is architected from the ground up for 10-Gig performance, and Extreme has the edge technology, security architecture, and management software to match.

Cisco 2811 Integrated Services Router won best SOBO services consolidation:-
Cisco's 2800 Series ISRs represent the next step in network infrastructure evolution, packing conventional routing, firewalling, PoE switching, embedded VoIP call management, and voicemail into a single 1U modular unit. It takes precious little room in the network closet, and all of these services can be managed from a single GUI.

And finally Network Physics NetSensory NP-500 won best network analysis solution:-
NetSensory NP-500 is a tremendous tool for peering deep into the inner workings of the network. The amount of data collected is impressive, but even more interesting are the nearly limitless ways that data can be sliced and displayed. We were amazed at how much we could discover with just a few simple mouse clicks.

Posted at 09:51AM Jan 04, 2007 by Simon Bullen in Networking  |  Comments[0]

Cisco plans to separate router hardware, software

December 15, 2006 (Network World) -- Cisco over the next five years plans to radically change how it sells and delivers router and switch software, in part by making that software more virtualized and modular.

Cisco's intention is to decouple IOS software from the hardware it sells, which could let users add enhancements such as security or VoIP more quickly without having to reinstall IOS images on routers and switches. The vendor also plans to virtualize many of its network services and applications, which currently are tied to hardware-specific modules or appliances.

This shift would make network gear operate more like a virtualized server, running multiple operating systems and applications on top of a VMware-like layer, as opposed to a router with a closed operating system, in which applications are run on hardware-based blades and modules. Ultimately, these changes will make it less expensive to deploy and manage services that run on top of IP networks, such as security, VoIP and management features, Cisco says.

High-level details of the road map were delivered in a session at Cisco's C-Scape analyst conference last week in San Jose by Cliff Metzler, senior vice president of the company's Network Management Technology Group.

"The way we've sold software in the past is we've bolted it onto a piece of hardware, and we shipped [customers] the hardware," Metzler said. "We need more flexibility to allow customers to purchase software and to deploy it according to their terms."

IOS upgrades require a reinstall of the new software image on the router or switch -- which causes downtime -- or, "we say, not a problem, UPS will arrive soon, here's another blade" to run your new service or application, Metzler said. "This adds months to the deployment cycle, which is not good for customers or Cisco's business."

Because IOS code releases are a superset of features in previous versions, Metzler added, users must also go through lengthy testing processes to ensure new features don't interfere with existing network services. The most recent IOS release, for example (12.4(11)T), has 31 new features, ranging from intrusion-prevention system (IPS) and VPN upgrades, to VoIP, Border Gateway Protocol, load balancing and VoiceXML features.

"What's going to happen? What else was in this software image that I just loaded?" are common questions when upgrading IOS, Metzler said. "It's not a natural, graceful way to go through a software upgrade."

The evolution of Cisco's software model was first mentioned by CEO John Chambers in June at the company's North American customer event.

"More than half of our engineers are software engineers, yet we sell [software] like a hardware product," Chambers said.

Cisco is not starting from square one in its effort to transform its IOS and network software technology and business. Cisco already has a modular IOS version, IOS-XR, which runs on its carrier-class routers, and provides a more flexible and resilient system for routing and advance services.

Posted at 11:30AM Dec 18, 2006 by Simon Bullen in Networking  |  Comments[3]

low power - low cost - high performance

A comment on my last post talked about the power consumption of each Sup720 in a Cisco 6509 chassis being in the region of 350W and how a Force10 S2410 only draws 150W through its Fulcrum powered processor, reminded me of the other vendors taking the route of Force10 and producing a low cost 1U 20 port 10Gb switch.

Fulcrum has had their chip out for some time and it has been adopted by Force10 networks, the unique aspects of this chip is that it is only layer 2, it only does cut through switching rather than the standard store and forward giving about a 200ns latency and its low power consumption of 150mW/Gbps per active interface. So mow any vendor can purchase the evaluation platform from Fulcrum add their code and you have a very quick cost effective platform (these are retailing about $18,000 list)



Other vendors taking this approach are Fujitsu and Broadcom.

Fujitsu's switch chip embeds 20 high-bandwidth, full-duplex 10Gbps ports in a single, integrated 1,156-pin FCBGA package. The firm claims its 10Gbps serial ports provide two advantages. First, they eliminate the need for expensive off-chip XAUI to XFI SerDes, allowing direct connections to optical XFP modules on any port and helping to significantly reduce cost, latency, power consumption, and board space. Secondly, the ports reduce the routing overhead inherent in running 4 x 3.125G lanes.
A big difference between this switch and the offering from Force10 is you have the option to go either cut through or store and forward.



Both Force10 and Fujitsu offer CX4 and optical solutions, although in my opinion the cost of 20 optics at $2k each kills off the appeal of these platforms.

Posted at 12:01PM Dec 11, 2006 by Simon Bullen in Networking  |  Comments[3]

More 10Gb ethernet

Cisco launched the new 8 port line card for the 6509 chassis giving a total 10Gb port count of 64. This module supports pluggable optics to support distances up to 80km over single-mode fiber, 300m over multimode fiber, and 15m over copper. Both the 4 and 8 port models are interoperable with the Cisco Catalyst 6500 Series Supervisor Engine 720 and provide 40 Gbps connection to the switch fabric.
Unfortunately until Cisco adopts a smaller optic they are going to struggle to increase the port count much further.

In the mean time Force10 are stuffing 224 10Gb ports into one of their E1200 chassis
The 16-port high density 10 Gigabit Ethernet line card for the Force10 E-Series E600 and E1200 is based on Force10 TeraScale™ ASICs and E-Series Architecture and provides pluggable XFP optics supporting distances up to 80 Km.
Each line card slot supports 48 Gbps of throughput or a density of up to 56 line-rate, non-blocking 10 Gigabit Ethernet ports in a single chassis. Additionally, the TeraScale ASICs and E-Series Architecture enables predictable performance with QoS and access control lists (ACLs) enabled, jumbo frame support, with simultaneous full functionality L2 switching and L3 routing.

Posted at 03:42PM Dec 08, 2006 by Simon Bullen in Networking  |  Comments[1]

3Com to Part Ways With Partner Huawei

3Com Corp. said it plans to buy out Huawei Technologies Co.’s stake in a networking equipment joint venture the two companies set up in 2003. 3Com will pay $882 million for Huawei’s 49% share in the joint venture, which is based in Hong Kong but has most of its operations in Hangzhou, China. The deal still needs to be approved by the Chinese government, said 3Com, which took majority ownership of the joint venture last January.

Posted at 11:45AM Dec 04, 2006 by Simon Bullen in Networking  |  Comments[0]

Ethernet knows where it's going: 100 Gigabits

A special study group of the Institute of Electrical and Electronics Engineers Inc. (IEEE) last Thursday agreed on a target for the next generation of the ubiquitous data networking technology. The 100G-bit/sec. version of Ethernet will be 10 times faster than the current fastest type, 10-Gigabit Ethernet. But vendors as well as users represented in the group, including Lawrence Berkeley National Laboratories and cable operator Comcast Corp., see a need for that kind of speed down the road, said John D'Ambrosia, chair of the group. It will serve the needs of both enterprises and carriers, he said.

Ethernet was introduced more than 30 years ago and became popular as a 10M-bit/sec. system for enterprise LANs. Along the way, Fast Ethernet (100M bit/sec.), Gigabit Ethernet and 10-Gigabit Ethernet have been added. Because each was standardized, many vendors have been able to compete and prices have been driven down. New Ethernet versions began life aggregating streams of packets from lower-speed connections and, in turn, were later aggregated into fatter pipes that used the latest high speed.

Video, high-performance computing and the increasing demands of applications in data centers will require faster connections, said D'Ambrosia, who is also scientist of components technology at high-end Ethernet switch vendor Force10 Networks Inc. The study group was formed in July to decide what speed the IEEE should try to achieve in the next standard. Last Thursday at an IEEE meeting in Dallas, 100G bit/sec. achieved the required 75% vote within the study group.

Other possible speeds were considered, including 40G bit/sec., 80G bit/sec. and 120G bit/sec., but none of them garnered enough backing. The group weighed the time and effort required to achieve a speed against how well it would meet the needs that exist when it becomes available, D'Ambrosia said.

Between that vote and 100G bit/sec. Ethernet hitting the market, the IEEE needs to approve the formation of a working group that will then figure out how to achieve the higher speed. Judging from the development of earlier standards, D'Ambrosia said standard 100G bit/sec. Ethernet products are likely to become available in late 2009 or early 2010.

The challenges this time will be similar to those in the past, only harder, he said. Among them are heat and power requirements and enabling faster communication among the chips inside networking equipment. Like other steps up in Ethernet speed, it is likely to appear first in gear that uses optical fiber. Getting it to work over copper wires will be harder than ever, D'Ambrosia said, but he wouldn't rule it out. The cycle never ends, he said.

"This will never be the last higher speed study group," D'Ambrosia said. "We'll get this done and eventually there will be a push for another speed after this."

http://www.computerworld.com/

Posted at 03:37PM Nov 29, 2006 by Simon Bullen in Networking  |  Comments[0]

Nortel 5510 performance

The other week I had to deal with an issue regarding performance of a Nortel 5510-48t set up, I found this quite impossible to believe on the basis this switch is one of the best stand alone units I have ever tested, combined with the 40Gb stacking function this switch is as solid as it comes.

In fact I had such faith in the 5510 I bet the on site engineer a Christmas present if I was proved wrong, so they swapped a single 5510 out with another switch and doubled the performance, bugger I owe his a present !

It seems the customer was creating "packet" sizes of 100kb to 1MB, these are more like TCP payload units, but this caused large volumes of TCP retries on the servers and performance was suffering badly. An investigation into the 5510 reviled that the buffer size for ports could be increase to accommodate the larger frames in a more efficient manor but this had drawbacks, this is available from ver 4.0 onwards.

Basically the way Nortel does the buffering on the switch is in groups of 12 ports, changing the buffer size for one port effects the whole switch.

The normal setting allows each port to access 12% of the buffers per group of 12 ports

The high setting allows each port to access 30% of the buffers per group of 12 ports

The Max setting allows each port to access 100% of the buffers per group of 12 ports.

Nortel for obvious reasons set recommendations for the amount of ports that should be in use per block of 12 in order to stop customers creating further issues, but for this particular case I am using all 12 ports and have the buffer settings set to Max without incurring any issues. I will spread the severs out to 8 per 12 ports and retest but at the moment performance is excellent.

So my faith in the Nortel 5510 has been renewed, this is still a excellent switch.

Posted at 10:02AM Nov 28, 2006 by Simon Bullen in Networking  |  Comments[1]

Bluegene vs Solaris/Opteron/Ethernet

Durham University are running a Solaris 10 X86 cluster consisting of 264 nodes of Sun Fire X2100 server connected together via standard Nortel 5510-48t gigabit network switches running 5.0.3 firmware and a configuration that is optimized for larger frames.

In order to assess the performance of the network Durham ran the Gadget-3 test suite across half the cluster (256 cores) and achieved some interesting and impressive results compared to a 1024 IBM Bluegene cluster:-

The same application of Gadget-3, running on:-

Machine	Run Time	Communication Time
1024 Bluegene processors	16518 seconds	340 seconds
256 Opteron Cores & Solaris 10	17692 seconds	5200 seconds

Also comparing this Solaris cluster to an equivalent 256 Linux node cluster which is using a custom TCP/IP stack bypass driver, the performance is essentially neck and neck. This is very significant in performance terms, but also users report that the Solaris environment is also more stable.

The other main factor that needs to be taken into account between the IBM and Sun clusters are the initial cost and support charges. The Bluegene solution cost over 4 times the Sun solution, the yearly maintenance cost of the propitiatory IBM solution is vast while Durham makes the most of Nortel's lifetime warranty which is standard on the 5500 series switches keeping ongoing costs can to a minimum.

Posted at 02:51PM Nov 22, 2006 by Simon Bullen in Networking  |  Comments[3]

Long live ethernet.

At SC06 last week I had a very interesting chat with the guys on the Fulcrum stand, they produce chips that can support up to 480Gb of layer2 cut through traffic making it a very quick switch.

Back on Thursday Aug 03, 2006 I wrote a piece about how ethernet would catch and overtake IB, it seems that I am not the only one who thinks like this.

http://blogs.sun.com/sbullen/entry/infiniband_vs_ethernet

What was interesting at the show was the volume of 1U 10Gb ethernet switches, Fujuitsu and Force 10 where displaying switches while Fulcrum and Broadcom were displaying chips and sample switches, all were in the $18k list price region.

Great T-shirt guys, one that I will enjoy wearing !

Posted at 02:49PM Nov 20, 2006 by Simon Bullen in Networking  |  Comments[1]

Voltaire InfiniBand Added To Sun’s Grid Computing Solutions Portfolio

Agreement Makes InfiniBand Available to Sun Customers Worldwide; New Sun Grid Rack System for Scalable Storage Offering To Include Voltaire InfiniBand

November 14, 2006 – Voltaire, the worldwide leader in grid backbone solutions, today announced that the company has entered into a global agreement with Sun Microsystems that brings Voltaire’s products into Sun’s portfolio of technologies for building integrated high-performance computing systems. Included in the terms of the agreement are Voltaire’s InfiniBand and multi-service director-class switches, host channel adapters and software, which Sun will use in pre-designed, factory-integrated solutions. The combined solution enables customers to gain industry-leading performance and scalability for their clusters and grids.

One of Sun’s key factory-integrated offerings for high-performance computing is the Sun™ Grid Rack System. The Sun Grid Rack System is a ready-to-deploy system with Sun servers, networking options and grid-ready software, delivered in a Sun rack. With a flexible, open architecture design, Sun Grid Rack Systems simplify the deployment and adoption of grid computing architectures. Voltaire multi-service Grid Director™ switches offer integrated InfiniBand, Gigabit Ethernet and Fibre Channel connectivity in a single enclosure and are deployed successfully in many of the world’s largest supercomputers and grids.

Voltaire technology is also part of the new Sun Grid Rack System model designed for storage grids. The Sun Grid Rack System for Scalable Storage combines the innovative Sun Fire™ X4500 hybrid data server, Sun StorageTek™ arrays, Voltaire InfiniBand fabrics, and the Lustre cluster file system, to ensure great scalability for data access at remarkable price-performance levels.

“Sun and Voltaire share a goal of delivering the highest-performance, complete cluster and grid solutions to our customers,” said Bjorn Andersson, Director for HPC and Integrated Systems, Sun Microsystems. “The new Sun Grid Rack System for Scalable Storage is an example of our collaboration, and adds to the great momentum we have in the market place with integrated products. We are now pleased to provide more options for integrating InfiniBand technology with the products from Voltaire.”

“While Sun and Voltaire have collaborated for quite some time to deliver high performance clusters and grids to customers worldwide, we are very pleased to take the next step by announcing the global agreement,” said Mark Favreau, president and head of worldwide sales for Voltaire. “We look forward to building on our relationship with Sun to deliver ultra high performance, innovative, yet easy-to-deploy solutions to our customers.”

Sun and Voltaire Power Supercomputers at Leading HPC Institutions Worldwide

Clemson University, University of Cologne and Japan’s Tokyo Institute of Technology (Tokyo Tech) have selected the Sun solution with Voltaire InfiniBand to deliver industry-leading performance to their mission-critical applications.

“Clemson University chose a Sun Microsystems HPC system and Voltaire’s state-of-the-art InfiniBand solutions to build a world-class simulation facility for the automotive and aviation industries,” said James H. Leylek, Ph.D, Director, Advanced Computational Research Laboratory, Clemson University. “The solution will help address technical problem areas as diverse as computational aerodynamics, vehicle dynamics, acoustics, materials, manufacturing, electromagnetism, wireless communications and others.”

“We selected Sun Fire x64 servers and a Voltaire Grid Director InfiniBand-based switch to develop a 128-node cluster to run computational analysis,” said Professor Ulrich Lang, Director of the Center for Applied Informatics, University of Cologne. “By leveraging 10 Gbps InfiniBand as the server interconnect, we have built a high performance, scalable cluster that allows us to tackle complex calculations faster and more efficiently than ever before.”

Tokyo Tech selected Sun servers and Voltaire’s InfiniBand-based Grid Director switches to power Japan’s largest supercomputer, which is used for computational scientific research. The Tokyo Tech system is one of the top ten largest supercomputers in the world as ranked by the Top500 (www.top500.org) and was built using Sun Fire x64 (x86, 64-bit) servers with more than 10,000 AMD Opteron™ processor cores, connected by multiple Voltaire Grid Director ISR 9288 switches.

Voltaire solutions are Solaris-Ready and have been certified on the Solaris™ 10 Operating System (OS).

Voltaire products for Sun integrated solutions are available today through the Sun Customer Ready Systems program.

Posted at 08:52AM Nov 20, 2006 by Simon Bullen in Networking  |  Comments[0]

Voltaire Connects Ten Percent of Top500 with InfiniBand

Number of Top500 InfiniBand-based Supercomputers More Than Doubles Since June 2006; InfiniBand Deployments Surpass Myrinet

SC|06, TAMPA, Florida and BILLERICA, Mass. – November 14, 2006 – Voltaire, the worldwide leader in grid backbone solutions, today announced that the company’s InfiniBand-based Grid Backbone™ switching solutions are powering ten percent of the world’s fastest supercomputers as ranked on the 28th edition of the Top500 list. InfiniBand deployments on the list more than doubled since June 2006 and two-thirds of the InfiniBand clusters leverage Voltaire solutions.

Two of the top ten sites on the list, NASA’s Columbia supercomputer and the GSIC Center at Tokyo Institute of Technology are customers of Voltaire. Voltaire Grid Backbone solutions deliver industry-leading performance and networking efficiency to enable highly scalable clusters and distributed computing architectures. Voltaire switches and software are used by businesses and research institutions around the globe to build high-performance clusters and grids ranging from hundreds to thousands of nodes.

The Top500 list (www.top500.org), published twice a year prior to the Supercomputing Conference held this week in Tampa, Florida, and the International Supercomputer Conference held in June, ranks supercomputers worldwide according to their performance on the LINPACK benchmark. The number of InfiniBand-based supercomputers grew 173 percent from the November 2005 list and 128 percent from the June 2006 list and is largely fueled by new deployments of Voltaire solutions.

For the first time, InfiniBand usage surpassed proprietary Myrinet interconnect usage on the Top500. InfiniBand also captured share from lower performing Gigabit Ethernet, which saw an 18 percent decline since the June 2006 list.

Voltaire deployments on the Top500 span High Performance Computing (HPC) institutions worldwide in the government, academic, financial services and semiconductor sectors. The Voltaire InfiniBand-based clustered supercomputers on the list were delivered through the broadest representation of partners, including: HP, IBM, SGI, Sun, NEC, Dell, Appro, and LinuxNetworx.

“The doubling of InfiniBand deployments on the Top500 list is indicative of the rapidly growing use of InfiniBand in HPC and across the enterprise,” said Patrick Guay, senior vice president of marketing, Voltaire. “We are extremely pleased to grow our leadership as the leading provider of InfiniBand solutions and look forward to working with our customers to begin the move to petaflop-class clusters.”

Posted at 03:17PM Nov 15, 2006 by Simon Bullen in Networking  |  Comments[0]

Cisco's profit and revenue surge ahead.

Cisco kept growing fast in its fiscal first quarter, posting revenue of $8.2 billion and net income of $1.6 billion or 26 cents per share.

Excluding certain items, the dominant networking vendor saw earnings per share of 31 cents, beating the consensus estimate of 29 cents from analysts polled by Thomson Financial. The analysts had predicted Cisco's revenue for the quarter, which ended Oct. 28, would be just under $7.9 billion.

As the company expands from its core switching and routing business into new areas such as security, telephony and videoconferencing, it has hit the mark in many cases and promised strong profit growth over the remainder of this decade. Results for the first quarter didn't disappoint, as revenue grew almost 25% and net income shot up 27.5% from the same quarter last year.

MORE

Posted at 10:02AM Nov 09, 2006 by Simon Bullen in Networking  |  Comments[0]