Comcast's pricey DOCSIS 3.0

Comcast has finally started its rollout of Docsis 3.0 in the northwest. Comcast will be offering this service in Seattle, Spokane, and surrounding areas in Washington, and Eugene in Oregon. How about the subscription cost? I must say that Comcast haas lived to its standards, it is extremely pricey. Here are the subscription costs I got from an article in Lightreading:


For its initial wideband service deployments, Comcast is leading off with an "Extreme 50" tier that offers bursts of up to 50 Mbit/s downstream and 10 Mbit/s upstream for $139.95 per month. The "Ultra" tier sells for $62.95 per month, offering 22 Mbit/s down by 5 Mbit/s upstream.

Comcast is coupling that with a business-class Wideband package (50 Mbit/s down by 10 Mbit/s up) for $189.95 that bundles in firewall services, static IP addresses, 24/7 customer support, and a suite of software from Microsoft


$150 a month (with taxes) for 50 Mbps downstream does seem exorbitant to me. Compare this to Korea/ Japan where customers pay around $40 a month for Ethernet PON (EPON) based broadband access; and you will feel that Comcast is ripping our wallet.

Whitepaper from Arista on 10 GigE penetration in SANs

Arista Networks released a joint whitepaper with Intel on 10 GigE performance technologies and penetration of 10 Gig Ethernet into SANs using the iSCSI protocol. The whitepaper is available here. The whitepaper discusses a bunch of Intel technologies for 10 GigE such as I/O Acceleration Technology (IOAT) and Virtual Machine Data Queues (VMDq). Also mentioned is how iSCSI is helping 10 GigE replace Fiber Channels in SANs.

Arista Networks also has a new CEO today, Jayshree Ullal, a high-profile Cisco senior VP. Also joining is Andy Bechtolsheim as Chief Development Officer. Andy is one of the founders of Sun Microsystems, which he re-joined in 2003 on the acquisition of Kealia. Arista vision is to enter the cloud computing market in which multiple services and applications are deployed in a central network and may be easily availed from multiple devices and end-points.

Neterion's whitepaper on IO QoS

Neterion has an excellent whitepaper on IOQoS, a technology that enables multiple virtual channels to have independent QoS parameters while sharing the same Neterion 1- GigE adapter. OpenSolaris has a similar project, Crossbow which lets virtual streams have independent QoS parameters while sharing the same 10 GigE interface.

Encryption at the speed of 100 Gbps

According to a Lightreading report, Telcordia is trying to bring optical security at 100 Gbps using Photonic Layer Security (PLS). PLS changes an optical signals' phase by different amounts at different points, and then applying a similar iteration on a combined optical signal.

That mixes things up, but the number of phase changes involved is small enough to be cracked by trying enough possible permutations. (There's a limit to the number of usable codes that can be applied to that step.) So, PLS calls for multiple signals to get combined, with a second encoding iteration done on the aggregated stream, again using phase shifts. The decoding can be done only with the PLS key.

For details on PLS, please refer to the Journal of Optical Networking (JON) article available here.

Notes from Intel Developers Forum

Intel Developer Forum (IDF) was held 19th-21st August last week. Intel seems to be focussed on three key markets: (i) Mobile Internet Devices (MIDs) (ii) Converged internet and multimedia, and (iii) High-end enterprise solutions. Intel is targeting these markets with the following processors respectively (i)Intel Atom which is sized smaller than a quarter coin but has as many transistors as the Pentium IV, (ii) Intel Media Processor CE 3100, and (iii) Intel Nehalem which is Intel's first foray into a NUMA architecture with high memory bandwidth using the Intel Quickpath Interconnect (QPI) technology.

The plenary talks revolved around the above. Out of a gamut of applications and gadgets talked about, I found two interesting. The first is Gypsii, an unique application combining the mobile computing and social networking. On a mobile device, such as your IPhone, you can locate all your friends on a map, and instantly communicate with them like hooking up for lunch. The other is a TV internet widget jointly developed b Intel and Yahoo (see press release here. With this widget, you will have a toolbar at the bottom of your television screen with which you can check your mail, stocks, weather, news, and what not.

Many of the technical sessions were based on parallel computing and the Nehalem architecture. Nehalem seems to have improved branch prediction, better unaligned cache handling, improved store and load performance, besides significantly higher interconnect bandwidth which should help in faster memory access and better I/O. That seems to indicate that Nehalem will have great 10 GigE network I/O, so it will be fun to do some performance characterization and analysis on a Nehalem box. Besides there was a 2 hour session on the Intel Advanced Vector Extensions (AVX) ins which is expected in 2010. AVX will be operating on 256-bit registers allowing for increased vectorization and 256-bit add, multiply and shuffle operations.

All presentation materials from IDF are now publicly available here.

Google policymaker argues about owning your own fiber

Here is an interesrting article from Google about our need to reduce the dependence on ISPs to deploy advanced fiber based broadband access services such as PONs. This coincides with Google chief Internet evangelist Vint Cerf criticizing ISPs for not deploying PON like infrastructure in the US (See article here). Google has also bought considerable mobile spectrum recently sold in the US.

These show Google's increasing efforts to reduce its dependence on ISPs to supply services to end-users. Interesting Google's blog refers to an "open access" like model (discussed earlier in this forum), in which people own the fiber connections to their home, and ISP's bid for the right to provide services to them. The blog draws an interesting analogy to cars which were first rented by people (like cabs), but were eventually bought by masses; and computers which were first shared in a mainframe, but are now owned by everyone.

Brocade buys Foundry

Brocade announced yesterday its intent to buy Foundry networks for $3 billion, a considerable premium over Foundry's current market cap. This deal is interesting in a variety of ways, but it definitely points towards the importance of Ethernet, particularly Fiber Channel over Ethernet (FCoE) in the future. And in many ways it could be the turning point of the ethernet revolution in the data center.

Technically the deal makes sense. Brocade is a champion of fiber channel. But with FCoE, it is clear that Brocade will be threatened by Ethernet based vendors taking away a slice of backend switches. With Foundry, Brocade can offer a product line complete with technologies over stages of evolution: from fiber channel, to ethernet, to FCoE. It also poses a challenge to Cisco's emergence in the data center.

While this deal sounds good, I have been always skeptical of the execution of the deal when two equal sized companies merge. It just seems difficult to bring two corporate cultures together when neither one can dominate given that the size of the two is almost equal. One prominent example which has not worked till now is the AMD - ATI tech merger, the combined company's net worth being $4 billion now compared to the $5.5 billion for which ATI Tech was initially bought. No it will be interesting to see how things move forward after the deal.

Chelsio now supported on VMWare ESX 3.5

If you missed this press release from VMWare- Chelsio's 10 GigE family is now supported on ESX 3.5.
Also, Centillium has been bought by TranSwitch.

PMC-Sierra White Paper on 10 GigE

PMC-Sierra has released a white paper on their 10 Gig reference design. While the paper does not talk about much technical details on 10 GigE, it does try to motivate 10 GigE the obvious way, IPTV. It talks about the growing demand for HDTV, and how people with multiple set top boxes would require bandwidth that can be provided by 10 GigE solutions. Nothing much new, but may be interesting read for those in this field. Paper may be downloaded from http://www.pmc-sierra.com/ftth-pon (Registration required).

Welcome to our new readers

We are now using Google Adwords to advertise this blog, and we welcome new readers who are discovering our blog via this channel. A quick introduction to this blog, we have discussed developments in Ethernet and related technologies for close to three years now. Feel free to contribute to current posts.

We have also an excellent search feature now, again courtesy google. We regularly refine and update this, and we recommend you to use this to find other posts and lively discussions on this blog.

Happy reading.

Comcast reports testing 100 GigE

Comcast today issued a press release stating that it had successfully tested 100 GigE over its backbone network between Philadelphia and McLean, VA using Cisco's 100 GigE interface for its CRS-1 platform. The release also mentions that Comcast followed the specifications for the IEEE 802.3ba standard for 100 Gig Ethernet.

While this makes Comcast one of the first service providers to test a 100 Gbps point-to-point tunnel, I would like to remind my readers that this itself is not a major technological achievement, however colorful the news may read. Dense Wavelength Division Multiplexing (DWDM) has existed in backbone networks and the optical world for several years now. So has the Cisco CRS-1 which can route multiple OC-192 (equivalent of 10 Gbps in the optical world) and OC-768 (40 Gbps). In fact Cisco claims that its CRS-1 can route upto 92 Terabits of traffic. I was not aware of a 100 Gbps interface to the CRS-1 before, but I presume that Cisco has it now, at least in prototype.

So carrying 10 optical channels across a backbone network is not something new. What's new about this I presume is that the 10 optical channels are delivered over a single interface which follows the 100 GigE IEEE 802.3 ah standard. Moreover 100 GigE has to my knowledge not been demonstrated before on a WAN, although there are previous studies from Lucent and AT&T showing 100 GigE on a LAN. For those unaware, the IEEE 802.3ba draft deals with how to handle several 10 GigE channels on a Ethernet interface. I am curious about which vendor Comcast worked with for the 100 GigE end interface. Opnext and Hitachi are the two companies having EA-DFB Lasers that can support 100 GigE according to this article.

EPON supporting DOCSIS from Salira

Apologies to my readers for no updates over the last month-and-a-half, I was busy traveling. There have not been any major updates in our field, except for the news of Salira releasing DOCSIS support for its EPONs.

The idea is to use EPONs as a feeder network for Cable TV networks. DOCSIS 3.0 which has been standardized recently and is still under trial deployment by Comcast and others and support downstream bandwidth is excess of 12 Mbps. PONS seem the natural choice to act as feeder networks.

However, this idea is not new. CATV networks already use fiber as a feeder till the last mile where copper takes over. EPONs have also been suggested earlier as feeder networks to VDSL, whose reach is much shorter. So much is to be seen on how Salira plans to market this.

Article on EPON

Here is a recent article on Ethernet PON and 10 Gig EPON. The author discusses the two technologies, deployments, and market share of the two companies. Also discussed is the market demand for 10 Gig EPON technology. The author rightly points out that the utility of 10 Gig EPON depends on convergence of business and residential services, and growth in demand for HDTV over IP.

Cisco Catalyst 4900M switch

Here is an excellent article about the Cisco Catalyst 4900 M switch which was released last January.

Broadcom backing Multi-link distribution (MLD) for 40 GigE

According to this article, Nick Ilyadis, Broadcom CTO for its Ethernet Networking group mentioned that Broadcom will be supporting the multi-link distribution (MLD) proposal for 40 GigE which is expected to be the solution adopted by the IEEE high speed ($0 Gig and 100 Gig) task force. A brief overview of MLD is available from the IEEE 802.3ba public area here.

I have not been following up much on the activities of IEEE 802.3ba, could anyone from the committee or outside could comment on this?

News from Chelsio and Neterion

Lots of news coming in from Chelsio today. Chelsio announced that its 10 GigE card will be supported by Citrix XenServer which is based on the open source Xen virtualization code base. Moreover Chelsio also announced SGI and HP as its customers. HP will supply Chelsio's T3 10GbE adapter for its Integrity server line.

It also seems (according to this article) that Chelsio will be coming out targeting Fiber Channel over Ethernet (FCoE). FCoE is extremely latency sensitive, and Chelsio may have an important advantage in this space because of its hardware supported TCP stack, which may be able to provide more guaranteed latencies. (Edit: As a reader correctly pointed out, FCoE does not use the TCP stack, so my argument of Chelsio having an advantage for FCoE is incorrect. Thanks for pointing this out.)

Neterion, the other prominent 10 GigE vendor also announced that its driver will be released part of Windows Server 2008.

It seems to be time that 10 Gig Ethernet is entering a high volume market. 10 GigE adapters are being increasingly deployed in servers, and they seem to be just ready to enter the storage market. Technologies such as Virtualization and IOV will help increase the volume of 10 GigE. It will be interesting to see how these two startups (which have actually been in business for over 5 years now), evolve over the course of the next two years.

Chelsio releases TOE support for Windows Chimney

Found about this a little late... Chelsio released a TCP/IP Offload Engine (ToE) driver for Windows Chimney, Microsoft's network stack architecture which allows offload of the TCP stack to supported ToE hardware. Here is the press release.

Another nice article about adoption of ToE in the storage world is here.

10 GigE Storage Alliance formed

In an interesting development, Nimbus Data Systems, in partnership with Arastra, Force10 Networks, Fujitsu, Fulcrum Microsystems, Mellanox Technologies, Neterion, and NetXen, have formed the 10 Gigabit Ethernet Storage Alliance. The purpose for now is to promote industry awareness of consolidating 10 Gig Ethernet in storage platforms. Although the press release mentions various advantages of 10 Gig Ethernet consolidation such as reducing storage network infrastructure cost by 30-75% compared to 4Gb Fibre Channel, having more interoperability that 10 GigE provides, it doesn't mention technologies such as FCoE which other vendors such as Cisco are promoting. I wonder if they plan to use any competing protocols for 10 Gig E in the storage space. Any comments on this will be helpful.

1 Gig EPON reference designs available from PMC-Sierra

PMC Sierra now has reference designs for its IEEE 802.3av (syet to be standardized) compliant 10 Gig EPON with a capacity to handle 128 ONUs and the capability to simultaneously operate with 1 Gig EPONs and have backward potability of Dynamic Bandwidth Assignment (DBA) algorithms. See this link for details.

Big waves in FCoE

Fiber Channel over Ethernet (FCoE) is a technology which aims at I/O consolidation over Ethernet (mainly 10 Gig Ethernet) by encapsulating Fiber Channel frames over Ethernet. FCoE's success will ensure 10 Gig Ethernet's penetration into the Storage Area Network (SAN) market which has been till now mainly dominated by Fiber Channel and iSCSI. The main challenges to FCoE are (i) Encapsulating Fiber Channel frames over Ethernet (ii) Turning Ethernet into a lossless medium just like fiber channel is and (iii) Extending the Ethernet concept of MAC to fiber channel. To understand more about the technical Details about FCoE, I would recommend the following whitepaper.

FCoE is currently undergoing standardization at the T11 standardization body and a standard is expected towards the latter half of this year.

Today there was a lot of activity in FCoE. Cisco, Intel, and Brocade, Emulex, and Netapp jointly made product announcements in Storage World Conference. Cisco decided to completely acquire Nuova Systems, a startup that it had invested earlier in, and which has been working on FCoE switches for the last two years. Cisco also announced its Nexus 5000 switch which supports FCoE, a collaboration between Cisco and Nuova, and said it will have a price of $900 per port.

On the adapter side of things, Emulex introduced the LightPulse LP21000 platform of converged network adapters that support FCoE, Qlogic launched its 8000 series of FCoE-compliant converged network adapters, and Intel said its 10-Gigabit Ethernet adapters now all support FCoE while Brocade will be partnering Intel to deliver FCoE adapters using Intel 10 Gig cards.

It will be interesting to see how the FCoE market grows. It seems still a little premature to make a prediction since the 10 Gig Ethernet adapter market is still lacks volume penetration primarily owing to high costs. Similarly it will be exciting to see how FCoE contributes to the general consolidation market that technologies such as virtualization are driving.

Ethernet Expo and Provider Backbone Transport

LightReading is showcasting PBT in Ethernet Expo at London, April 14-16 this year, and had a nice summary article about PBT being in the spotlight today. Here is the link to the news article.

April Fools Pranks

Here is a story from Lightreading about Google going nuclear.

And here is a video of a prank played on Sun Microsystems' CEO Jonathan Schwartz. Well done.




Enjoy.

The question of net neutrality

Three days ago, Comcast announced that it is reversing its policy of blocking peer to peer (P2P) traffic selectively during peek hours of traffic. Although Comcast had never announced this as a stated policy, numerous studies in last year or two had showed P2P performing substantially poor over Comcast compared to competing ISPs. So the question of Internet neutrality has become an immensely debated subject over the last year. Many ISPs are actively lobbing Congress against Internet neutrality, while there is at least one prominent group, SaveTheInternet.com for net neutrality.

The reason I take up this for discussion in this forum is that it is actively tied to the idea of open access, which has been discussed here in past, see here and here for more details.

While I don't agree with Comcast in this case since I think it illegally manipulated traffic without making its customers aware of it, I don't claim to be a proponent of net neutrality either. I think that there is a legitimate demand for Quality of Service (QoS) applications such as video and voice of the internet, and the same cannot be delivered by carriers and content providers unless there is a sound business and revenue model to deliver the same.

Currently we lack a good availability of QoS because of what an erstwhile colleague of mine correctly described as "fingerpointing". In short, Content Providers blame ISPs for not having QoS guaranteed pipes, because of which they cannot deliver enhanced QoS applications. ISPs blame users for not willing to pay for such services because of which there is no revenue model to deploy infrastructure to deliver QoS. And users blame content providers for not providing such QoS services for which they would be willing to subscribe to.

Not sure of how to get out of this, but I think there is a need to fast express lanes in the Internet for advanced services. After all, no one complains when they pay a toll to use a bridge on a highway. Similarly in many cities of the world, there are tolls to use express roads in certain hours. I am sure that a similar approach will work well in the internet.

Any thoughts, please post your comments here.

Virtualization and Networking

I wrote a detailed article at my Sun Microsystems blog on how network I/O works with virtualization. It discusses the possible solutions: binary address translations, paravirtualization, and hardware assisted virtualization. Here is the link:

http://blogs.sun.com/hotnets/

Getting more out of your network adapter

As we go to higher bit-rates, it is typically difficult to achieve lie rate out-of box performance, because of many potential points for a bottleneck. Here is an article that discusses various strategies to get higher network I/O throughput.

Neterion releases IOV compliant 10 Gig Ethernet interface

Neterion today released its X3100 Ethernet adapters supporting the recently standardized Single-Root IO Virtualization (SR-IOV). Neterion released the card today at VMWorld in Cannes, France. SR-IOV is a PCI-SIG workgroup extension to PCIe that allows a single adapter to truly behave like multiple physical adapters in a virtualized server.

It appears that Neterion is placing high bets on virtualization. This seems interesting given that virtualization is likely to increase the demand for high-speed network adapters. It would be interesting to know how much overhead SR-IOV reduces compared to previous NICs of Neterion.

Multi-root I/O virtualization standard is also near. Can someone give some insite into the IOV standardization process?