Mesh Without Wires

December 21, 2010

Layer 2 vs Layer 3 in Wireless Mesh: Do You Have to Choose?

Filed under: Technology,Wireless,Wireless backhaul,Wireless Mesh — kseniacoffman @ 12:37 pm

There are many questions regarding Firetide’s mesh technology, and this posts attempts to address some of them. As you will see from the post, Firetide’s mesh is neither Layer 2 nor Layer 3, but a hybrid, unique to the industry.

Layer 2 / Layer 3 Hybrid Approach

Layer 2 vs Layer 3 benefits and trade-offs have been a topic of discussion for a couple of decades as both approaches have their benefits and drawbacks. Firetide offers a L2/L3 hybrid approach, which leverages the positives from both technologies and does away with the shortcomings, such as Layer 2’s lack of scalability and Layer 3’s latency and chattiness.

Firetide’s technology looks as a Layer 2 switch to outside world. Internally Firetide uses a Layer 3 approach to deliver the packets from an injection point of the mesh to an exit point. In doing so the packets can be load-balanced and re-routed based on advanced metrics maintained within the mesh. The hybrid approach makes the wireless network more scalable as well as enhances its performance.

Distributed Wireless Ethernet Switch

Firetide infrastructure mesh is built out of Firetide mesh nodes, which, combined into a network topology (mesh, point-to-point, point-to-multipoint, or a combination), form a “distributed wireless Ethernet switch.” The mesh nodes (in our terminology) are neither access points nor routers. Hence we call them nodes.

The term “distributed wireless Ethernet switch” means that the features provided are equivalent to a traditional wired Ethernet switch. In a wired Ethernet switch, Ethernet interfaces are provided via line cards on a chassis based model. The central switching is done via a central card that switches Ethernet packets from one line card port to another line card port. Firetide follows a similar approach. The line card equivalent of Firetide’s product offering is the mesh node, while the central switch is the radio medium itself. The nodes may be a mile to several miles apart.

The simplicity of Ethernet switch is illustrated the point that a 100-port Ethernet switch needs one single management IP address. In the same way, a 100-port Firetide mesh (formed with 25 indoor mesh nodes, for example) will have a single management IP address. The distributed Ethernet switch architecture is patented technology of Firetide and provides:

  • Seamless transport of IPV4 and IPV6 packets
  • Load balancing across the entire mesh
  • End-to-end encapsulation (on top of encryption) for added security, reliable handling of multicast traffic, and no introduced jitter
  • Simple installation and expansion of a mesh network. With a ‘routed mesh,’ this would require a complex configuration whereas each node has unique user-described IP address etc.

Firetide is the only provider with a distributed Ethernet switch-based infrastructure mesh. Most other wireless mesh in the market is a simple extension of access point-based technology, although some Layer 3 mesh also exists.

Flow-based Routing

The stitching of Layer 2 and Layer 3 is done on a flow-based model. The flows are Ethernet flows that are maintained within the mesh domain. The flows are then wirelessly switched across the distributed radio infrastructure.

Firetide built a protocol from the ground up with wireless in mind. Our flow-based routing algorithm creates a unique tunnel by encapsulating each packet with a flow instruction header specific to each stream; the header defines the least cost route for that specific packet from its source to its destination. Since flow instruction headers are used to determine the destination, the packets do not undergo deep packet inspection at each hop. In this way, we avoid extra overhead created by the millions of pings involved in resetting routing tables. Reduced overhead means more throughput for the user traffic.

These flows also help maintain unique, synchronous streams end-to-end. By maintaining the packets in these non-interleaved tunnels, the mesh makes sure that the packets arrive at the destination on time, in order, and with no introduced jitter (variation of latency).

Firetide mesh networks use load balancing across multiple paths for maximum network capacity. The mesh nodes manage the traffic across the network and can intelligently decide along which of 3 predefined flows a packet should be sent.

Performance

Firetide offers industry’s only wireless mesh that provides 100-150 Mbps of UDP payload sustained over multiple hops in outdoor deployments (real throughput, not theoretical data rate). When deployed in point-to-point mode, through radio bonding, we can provide real-world throughput of 200-300 Mbps outdoors.

What Firetide Mesh is Not

Even though Firetide mesh looks to the outside world as a Layer 2 switch, it does not perform “Layer 2 switching,” nor is it based on bridging two Wi-Fi radios together in WDS mode. On the switching side, Firetide mesh uses cut-through forwarding, as opposed to store-and-forward methodology. (Here’s a good definition from Cisco: “Whereas a store-and-forward switch makes a forwarding decision on a data packet after it has received the whole frame and checked its integrity, a cut-through switch engages in the forwarding process soon after it has examined the destination MAC (DMAC) address of an incoming frame.” See more on the two approaches: Cut-Through and Store-and-Forward Ethernet Switching for Low-Latency Environments.) In other words, Firetide mesh protocol performs proper routing across multiple hops, with cut-through forwarding approach allowing for industry’s lowest latency in wireless mesh: sub 1 ms per hop.

More questions? Please let me know in comments or via Twitter.

For more discussion on mesh technology, see

By Ksenia Coffman – Connect with me on Twitter or LinkedIn.

6 Comments »

  1. Ksenia, thank you for your post.

    It will sound like I am an employee of the competitor but this is just my practical experience with mesh equipment. While I do not have any experience with Firetide so please feel free to correct me where needed. I am always looking for the best mesh product and open to criticism when wrong. I have experience with “so called mesh” products like Ruckus, Motorola, and limited Cisco and true mesh products like Tropos, Bel Air and Strix Systems. In my personal experience Strix Systems solution is head and shoulders above everybody else.

    “…such as Layer 2′s lack of scalability…” —> While this is generally true, Strix Systems handles scalability with ease. You can put thousands of nodes (multiradio nodes) in a single subnet with no reduction in performance. Easy to set up, easy to scale. “Once upon a time” you had to plan for putting more subnets in a cloud. While you had to be more careful in design (the wireless mesh happened only on subnet per subnet basis) you could still do fast roaming from subnet to subnet via VLAN. But this is “pase” now, now you just put every node in the same subnet. Strix Systems uses proprietary L2 protocol for multiradio real time switched performance. Similar to Firetide this means that Strix Systems does not discriminate between IPv4, IPv6, Apple Talk or any other higher layer protocol. Multicast, end to end encapsulation, load balancing are Strix Systems features too.

    While it is a new information to me that Firetide is using L2/L3 approach the architecture sounds logical.

    “Firetide is the only provider with a distributed Ethernet switch-based infrastructure mesh. Most other wireless mesh in the market is a simple extension of access point-based technology, although some Layer 3 mesh also exists” —> Agreed. True mesh should be switch like infrastructure mesh. But Firetide being the only one. No. Firetide should remember that not so long ago Strix and Firetide had head to head comparison done by Iometrix: http://www.lightreading.com/document.asp?doc_id=97244. Granted that Firetide used just single radio indoor unist but you should still remember and honour the competition.

    “Firetide offers industry’s only wireless mesh that provides 100-150 Mbps of UDP payload sustained over multiple hops in outdoor deployments (real throughput, not theoretical data rate). When deployed in point-to-point mode, through radio bonding, we can provide real-world throughput of 200-300 Mbps outdoors” —> Also a bold statement. I can get more that 100 Mbps UDP over 10 hops with Strix and that is using 2×2 MIMO solution with aggregated latency between 10-30 milliseconds (on the wireless side. Core and and client latency are not included in this number but would add around 40 milliseconds). I have yet to receive 3×3 version of the product so I can not comment on it.

    What I do not like/understand with Firetide is using “just” two radios per node approach. Please have in mind that I do not know your product line very well. But to keep the performance high trough multiple hops (We can’t forget that wifi is a half duplex technology) you can not sustain both wireless backhaul and client access with just two radios. That would mean that “client” radio is working on one frequency (usually 2,4) while the backhaul works on one (both ingress and egress on one radio) as well (usually in 5 GHz range). And if you have multiple nodes connecting to each other on one frequency… well we WLAN folks know what that means, right?

    So in order to keep the performance I guess you use separate products for backhaul and for client access, right? How do you connect the client product to the backhaul product then?

    Strix Systems on other hand uses up to 6 radios per node (so the ingress and the egress always happens on the separate radios). That architecture allows for amazing performance trough multiple hops. The throughput degradation is less than half on the 10th hop and the latency between hop is 1-5 milliseconds. Using the proprietary protocol they developed a fast roaming product as well. It can roam from node to node with more that 120 km/h in a city environment and more than 300 km/h in straight line (like in trains) while keeping the high throughput and latency low.

    Thanks again for the debate!

    Regards,

    Gregor Vucajnk, Europe
    twitter.com/gregorvucajnk

    Comment by Gregor Vucajnk — December 21, 2010 @ 2:34 pm | Reply

  2. Thanks for the comment, Gregor. Honestly, I have not kept up with Strix these past couple of years. I believe they had been acquired by a company out of India, and then all but disappeared from the US market. Perhaps they have a stronger presence in EMEA and Asia. Although we don’t see them in deals there either, otherwise I would’ve heard. Even with this new info, we are still ahead – we shipped 3X3 MIMO over a year ago, hence probably the higher performance – to 150 Mbps, even with “just” 2-radio product. :-)

    Regarding the number of radios on a Firetide mesh node – yes, these are dual-radio mesh nodes, but both radios are used for mesh backhaul; there’s no client (Wi-Fi access) radio in a Firetide mesh node. While Firetide mesh uses 802.11 radios, our mesh does not conform to 802.11 standards. We modified the MAC layer, and are also running a proprietary routing protocol within the mesh. From the user perspective, however, Firetide mesh is 100% Ethernet compliant, which allows for a seamless integration with wired Ethernet networks and IP-based devices.

    To enable 802.11 standard communications (Wi-Fi), a Firetide mesh node needs to be co-located with an access point (Firetide’s own or 3rd party). To the Firetide mesh, the AP would be a ‘end device’ – same as plugging an AP into an RJ-45. In a similar fashion, other IP devices can plug into the mesh nodes – IP cameras, VoIP phones, etc. (This had set Firetide apart from other Wi-Fi mesh vendors in the past, which integrated Wi-Fi access point and backhaul radio(s) into a single enclosure. Our approach from the start had been to separate Wi-Fi access from mesh backhaul, for which we had been derided by other vendors :-) But at the end of the day, it’s proven to be the right approach, since we are all about IP infrastructure, and our mesh is much more than Wi-Fi access only).

    Regarding 6 radios vs 2 radios – this is a matter of implementation and architecture. In our view, if you can do the job with 2 radios, why use 6? The more radios, the higher the cost, complexity, and weight. I’ve not seen the Strix 6-radio product, but I know that Belair has a 4-radio solution: it weighs a ton and costs over $10,000 (don’t quote me on the number, but should be about right). And it still struggles with video, despite the four radios. Hence we don’t see BelAir that much in public safety and security, although they had made a concerted effort a few years back.

    You also mention that with Strix “throughput degradation is less than half on the 10th hop.” So despite the 6 radios, the mesh is still losing nearly 50% of throughput at 10 hops? Thanks to Firetide’s radio switching (receiving on one, and sending on the other) and the flow-based routing protocol, throughput degradation over multiple hops should be non-existent, if radios are operating on different channels, antennas are properly cross polarized to avoid self interference, etc. This is not always the case, due to interference or heavy channel utilization on one of the links “in the chain.” But in a ‘clear’ environment (minimal interference & lightly utilized channels), there won’t be any performance degradation attributable to the Firetide equipment.

    Hope this answers some of the questions you had. Thanks for stopping by and commenting!

    Comment by kseniacoffman — December 21, 2010 @ 4:45 pm | Reply

  3. Thanks for the reply Ksenia.

    Yes. It is true. They have been acquired by an Indian company and yes they do have stronger presence in EMEA and Asia.

    Sorry if I am not impressed with the “we had the 3×3 a year ago”. Every technology needs to mature and 802.11n is no different. Today the chipset (atheros, intel,.., pick one) are still bugy. So what you need to have is a low level coding to iron out that bugs and that takes time. We have to remember that the chips are standard based and not meant/produced for mesh networking in the sense you and I see it. Even more, the difference between 2×2 and 3×3 is not great at this moment. When using the 40 MHz wide channel the only difference is in the antenna, so you would usually have one MIMO connected to the vertical polarized antena, one to the horizontal and one to the 45 degrees slant. By using proper antennas (45 degrees cross polarized – like they use them in GSM/UMTS world) or circular polarized antennas (one MIMO on the left hand circular polarization, other on the right hand) there is really not much difference between the 2×2 and 3×3.

    I completely understand about modifying the MAC layer. Wi-Fi has no standard for mesh protocol. They are working on the 802.11s but that is not what two of us are talking about, right? Strix handles ethernet and IP in a similar fashion you are describing.

    OK. Now I understand how you can away with using just two radios. You dedicate them both on backhaul, one ingress one egress. So what is better. Having client connect radios in the same enclosure or in the separate? I guess it comes down to the business hand. I can see how you have great success with surveillance, CCTV, etc. So if you have a case for CCTV your product may be the right choice. As you have mentioned, two radios are more CAPEX friendly than 6 radios.

    I can also understand you being more than just a Wi-Fi product. Similary Strix System is technology agnostic. While they are predominately using WI-Fi, you can use 4,9 product line, Wimax or LTE cards with the same box, same system.

    6 radios vs 2… I agree on the logic. If you need two why build more. But this is a case for using omni or at least two 180 degree antennas. The wider antenna pattern you have the more noise you collect. By splitting the backhaul to more radios you can use more directional antennas. Also we in Europe are limited to 100 mW EIRP max on 2,4 and 200 mW to 1 W (depends on the UNII band) on 5 GHz. So you have to be extra careful when designing the mesh. More radios help.

    About having client access radio in the same enclosure or not… Again, depends on the business model. In my country you can not sell CCTV (we are not London :)). So I have to blanket cover as much of the area with as little nodes as possible so I can resell multiple applications to multiple clients (VoIP, video, data, real time applications..) OPEX is very important here. Having all in one box helps managing the system and installing the system.

    I have tested the new Bel Air myself. It is not so good. They have yet to master the mesh protocol. However with proper technical knowledge you can make it work (see Minneapolis). But is is hard to setup, harder to scale and pricy. But to answer you question, Strix box (the 6 radio box) is even bigger and heavier :). But you can still set it up on the public lighting pole with ease. It also uses integrated antennas witch is something I despise. Yes, it is easier to set up, you do not have to design for RF but that just calls for bad performance. As with choosing the right WiFi equipment, choosing the right antennas is very important (hey, I have a topic for my next blog :)).

    Your last paragraph is a little marketing colored :). One thing is lab environment where proper mesh protocol can allow almost zero throughput loss. Other thing is live, RF full environment. My experience derives from the dense (8-20 nodes per square kilometer) setup in a city environment. You can do what ever you want on the MAC layer, RF is stil on the L1!

    You havent commented on the fast roaming and low latency :).

    I would very much enjoy the next round of dog fight, not only between Strix and Firetide but other vendors as well.

    Comment by Gregor Vucajnk — December 22, 2010 @ 12:26 am | Reply

  4. Interesting point on being able to support WiMAX, LTE in the same Strix enclosure. We are focused on the mesh backhaul, and for us Wi-Fi is a separate function. But I do (occasionally) get comments on how come the Wi-Fi access point is a separate box. So it’s a matter of preference/conditioning and the application. Regarding the comment “But this is a case for using omni or at least two 180 degree antennas.” – it’s actually rare that we use omni antennas in mesh design; typically it’s a patch/panel or sector antenna per radio. In fact, as you mention you’d want to avoid omnis, as you’d be picking up noise from all directions. Omnis are reserved for specialized cases, such as infrastructure mobility installations on moving vehicles.

    Regarding throughput over multiple hops, the real-world performance of Firetide is nowhere near losing 50% of throughput at 10 hops. In Seoul’s Cheonggye Waterway project I covered in Network Design Considerations for Wireless Video Surveillance the starting bandwidth was 24 Mbps (this was done with our non-MIMO product), and at the end of 10 hops, the network was still delivering 20 Mbps. We have projects in the US with even lower throughput degradation (same throughput regardless of number of hops).

    Firetide latency is lowest in the industry (we believe): .9 ms per hop. We have a strong infrastructure mobility solution, which I covered in an earlier post: Technology Behind Wireless Infrastructure Mobility. In essence, one radio on a mobile node is used for data connectivity, and the other is used for scanning; when a stronger signal is detected from another fixed node, the radios switch. This allows us to maintain real-time video at high resolution. This technology is already being deployed for real-time video surveillance in Seoul Subway (SMRT) and Mumbai Metro. Seoul Subway network supports 20 Mbps of real-world throughput off and on trains moving at 50 mph (The throughput number is combined: i.e. 5 Mbps up and 15 Mbps down; this was designed with our non-MIMO mesh). The possible speed of the moving vehicle would depend on the density of the fixed deployment; the closer the fixed nodes the higher the speed (mph) the mesh network would be able to cope with. In Seoul Subway and Mumbai Metro examples distances between static nodes are about half a km.

    Gregor, it’s good to meet someone who agrees with us that most of the mesh out there is the “so-called mesh.” A general comment on Strix: yes, they used to be a formidable competitor around the time of the test you referenced (2005-2006), before Firetide had a dual-radio backhaul product. We came out with dual-radio mesh in 2006, and Strix had not been much of a threat since. Besides, they had blown through their funding by the time muni Wi-Fi market collapsed, and as I mentioned all but disappeared from our competitive landscape, that’s why I have not been tracking them.

    Comment by kseniacoffman — January 3, 2011 @ 3:21 pm | Reply

  5. An excellent blog post and even more excellent debate. Kudos.

    Devin Akin
    Big Nerd Guy
    Aerohive Networks

    Comment by Devin Akin — January 11, 2011 @ 2:29 pm | Reply

  6. I am genuinely thankful to the owner of this web site who has shared this wonderful article
    at at this place.

    Comment by Kent — December 7, 2014 @ 6:05 pm | Reply


RSS feed for comments on this post. TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Blog at WordPress.com.

%d bloggers like this: