Ring and star: who whom?

    When building networks, there are two competing topologies: a star (of different options) and a ring (of different options). The star has one advantage - low oversubscription. The star’s shortcomings are a complex structure, and accordingly, the complexity of operation and high cost. A star is a solution for access users gathered in one place: the classic network of a management building in an enterprise, but not always, even in this case. After all, the building's core network, in compliance with fire safety standards, will always be laid through two cable risers in different parts of the building and this is again a ring, not a star: one ring / optics cable through these two risers and two center of the building’s network / data center core.

    The choice of a specific solution topology depends on the object and its features. A ring is always more profitable for distributed networks, since it is very, very expensive and practically impossible to make a star on a large distributed network. Therefore, a ring topology is the optimal topology for large enterprises, processing plants, urban networks, networks of the whole country.


    Fig. 1. The control system module of the plant.

    All network equipment manufacturers can work according to the star topology, and only a limited number of manufacturers - HP, Huawei, Extreme, can work on the ring topology on switches with the distribution of virtual networks throughout the campus without the use of complex and expensive technologies such as MPLS / VPLS.

    For example, take a couple of typical simple “ring” objects: a stadium and an airport.


    Fig. 2. The network of the stadium.

    For these objects, the ring topology has the following advantages compared to a star:

    1. Installing and maintaining the ring topology is much easier. Installing and maintaining the ring topology is much simpler, since access devices through the ring immediately enter either the server farm or the kernel, and the star also suggests an intermediate level - aggregation of channels from the access level. The reason for the star’s aggregation level is simple: at the aggregation level, the port price is much lower than at the star’s core level, as well as for more flexible application of different policies.
    2. For modern networks, reservation of communication channels from the access level to the level of aggregation and / or core is critical. If in a star you make reservation of cable routes along different paths, then physically you get a ring, and logically - a star. But at the same time, for each switch on access, you will have to drag individual cable routes to the switching points, since the optics are always cut and welded with the entire bundle, and not in separate wires. Each welding of optics along the path of the movement of light in the fiber increases the loss of the budget of the optics, and as a result, reduces the distance of the optical communication channels. It should also be borne in mind that there will be several times more work on welding optics with star topology than with ring topology.
    3. For stadiums, airports and enterprise networks, there are always at least two separate physically separated networks. For a ring - it's simple, for a star in distributed objects - the reality is sad.
    4. Star convergence will always be worse, since there will always be loops of logical paths between access switches, except when two slot switches work as one in the kernel. The convergence of the ring topology is from 50 ms (one ring) to 200 ms (subrings are connected to the main ring).
    5. The star has difficulties with scaling: the addition of cable routes anywhere is the pulling of a new additional cable, for the ring - an additional coupling in the existing optical bundle.
    6. In the case of the ring, we remove from the core access the always necessary and planned uplink speed from the switches that are most loaded: that is, we pay for band growth as necessary. And in the case of a star - we pay immediately for everything and do not really use the bandwidth of the connected uplink channels.

    The last point I would like to consider in more detail. It should be noted that the server’s bottleneck will always be: since if we connect 28 gigabyte switches (more than enough for even the largest stadiums in the world) according to the star topology, then we will have 2 * 2 * overload when connecting to the core 20GB (server module connection speed to the rings) / 28 * 20GB (28 access switches, two 10 GB / s uplink each) = 1: 7. In the case of the stack, this will be 40GB (two server stacks) / 4 * 20GB (four access stacks) = 1: 2. But the oversubscription on access for the star will be 20GB (two uplinks of 10GB each) / 48 * 1GB (access ports) = 5:12 = 1: 2.4, and for the stack 20GB (stack width) / 7 * 48 * 1GB (seven access switches in one stack) = 5: 7 * 12 = 5:84 = 1:17. For a standard access network, congestion is allowed 1:20. As we see in the case of a star, low congestion on access causes a choke for the servers. To solve these problems in the core of the network for the topology of the star you need to install expensive modules (but the question is - why add them, if there are already enough server ports?), But if you need to reduce congestion on the stack, you just need to add optical converters. Below is a comparison table.
    TechnologyStackStarRemarks
    Access Subscription1:171: 2Valid 1:20
    Reduce Access OversubscriptionA pair of optical converters - The cost of each SFP + is low - 3K. But, given the experience of building previous stadiums, where in general access is 100 MB, this congestion will be more than enough.
    Reducing oversubscription when connecting servers - Server Modules in a Modular SwitchQuestion: if there are up to 50 servers in total, why connect more ports, half of which will not be involved even theoretically ?!

    But here progress intervened in the struggle between rings and stars: 40 GB and 100 GB interfaces appeared. For 95% of customers, these are simply unattainable in the near future bands of uplink communication channels for connecting access to the core. And as a result, for the same veins, without adding anything, but simply replacing the 10 GB transceiver with 40 GB or 100 GB, we increase the bandwidth by several times, and here the potential problem of re-subscribing the ring does not appear in principle. Although, again, the price of 40 GB and 100 GB of transceivers per 10 km is now very high, but in a year it will be the price of the current 10 GB transceivers.

    The largest examples of the use of ring technologies in conjunction with stacking technology in Asia are the city of Beijing, in Europe - the French railways. But the French railways simply had no choice after TR "died." And these are projects implemented on HP equipment (formerly 3Com equipment), which gives cause for reflection and practical application of the described ring design for data network designers.

    Also popular now: