Home > Articles > Operating Systems, Server > Solaris

  • Print
  • + Share This
Like this article? We recommend

The Sun Ray Environment Backnet Dilemma

Sun Ray appliances place, as previously mentioned, high demands on the quality of the backnet. These demands can be summarized as follows:

  • Sufficient bandwidth (application-dependent)

  • Low latency (less than 50 ms round-trip between Sun Ray server and DTU)

  • Ensured in-order packet delivery

  • Near error-free networks (less than 0.1 percent packet drop)

NOTE

Errors can occur, but they must be insignificant.

Ethernet network errors are usually caused by over utilization, misconfiguration, or physical defects on a network segment, causing packets to be dropped.

Over utilization causes collisions. The collision detect (CD) function of Ethernet causes communication to cease for a certain time frame (9 ms + dt). Collisions destroy, fragment, or misalign packets sent over the wire at the moment of collision. The CD function causes latency. In severely affected segments, communication becomes impossible (collision storms). Collisions are avoided by using good quality Ethernet switches that can operate in duplex mode, and are designed and implemented in such a way that they can handle the load caused by the network traffic.

Typical misconfigurations causing errors on the network are either ARP/RARP storms or duplicate IP addresses. Both causes can be determined by using network analysis tools.

Physical defects might be introduced by terminating equipment (hosts and other Ethernet-connected devices) or transfer equipment (switches and so on), but most likely, the defects stem from faulty cabling or connectors. The most common culprits are faulty patch cables. Physical errors are found by containment and measurement of the suspected equipment.

Sufficient Bandwidth

The trade-off is ensuring that there is sufficient bandwidth most of the time, or guaranteeing sufficient bandwidth all of the time. The latter is done by ensuring that the sum of all possible traffic never exceeds the capacity of the network. This is, however, very costly. Usually devices never operate at their potential peak load at once, so the capacity of a network designed this way is usually grossly oversized.

You can ensure adequate bandwidth by testing and calculating the actual peak loads, average loads, and minimum loads to deduce the capacity needed for the network components. This method can lead to a theoretical over subscription of the network. It is common to see over subscription numbers in the order of 2:1, 3:1 and 4:1. It is not uncommon to see numbers in the order of 10:1.

NOTE

In an oversubscribed network, the theoretical need for bandwidth (the sum of the maximum capacity of all network attached devices) supersedes the actual capacity of the network. It is common and quite safe to oversubscribe networks if proper investigation has been made. Oversubscription is done to satisfy the actual need for bandwidth, while keeping equipment costs down.

Example:

When dealing with Sun Ray appliances, we measured the actual need for bandwidth and found that usually a Sun Ray appliance has very low bandwidth demands, but can peak up to approximately 25 Mbps when running streaming video display programs. Based on this, we could say (bandwidth-wise) that if Sun Ray appliances are attached through 100 Mbps networks, and are all running heavy streaming video programs all the time, it would be safe to have an oversubscription rate of 3:1. This rate would guarantee that the Sun Ray appliances would work at all times. The other extreme occurs when users do not run streaming video at all. In such cases, the oversubscription rate could be 10:1 on a 10 Mbps network. A single Sun Ray appliance often uses less than 10 Mbps, and most of the time the appliance transmits user-generated traffic. These examples are not a recommendation, but examples of how you could reason. Measurements and risk assessments give the correct oversubscription rate for a particular configuration.

Low Latency

Latency can crudely be described as equipment handling time. In Sun Ray environments, low latency is a priority because the performance of the Sun Ray appliance, in effect, is dependent on the network performance. Latency is caused by collisions, as mentioned previously, but also inter-LAN links can cause latency, as can VLAN technologies.

When deploying Sun Ray appliances in a network, the "golden rule" is to make the architecture as uncomplicated as possible. This reasoning is seemingly the opposite of the mere thought of deploying Sun Ray appliances in a complex environment such as a MAN or a CityNet. Later, we will show that it is possible to combine MANs and low latency.

In-Order Packet Delivery

Because Sun Ray appliances use UDP as an information carrier, there is no way for the network to determine if the packets arrive in the correct order. This determination has to be handled in the application layer. For Sun Ray appliances, it is a design choice to not wait, but instead, consider out-of-order packets as lost packets. This design choice causes a problem when load sharing over multiple network trunks, as shown in FIGURE 1.

Figure 1FIGURE 1 Load Sharing Over Multiple Network Trunks

Sun Ray appliances can tolerate occasional out-of-order packets, but if this happens too frequently, these appliances might not work in that environment. In this theoretical model, Ethernet switches are set up as a load-sharing set. One packet can travel over the one interswitch trunk leg, and the next packet over the other leg, depending on the load situation for the particular devices involved.

The latency for the components depends on a fixed constant for the hardware, plus a variable based on the load of the device in question. The two legs are not dependent on each other. A packet leaving the server routed to one leg can arrive later than a packet leaving the server afterwards routed over the other leg. The Sun Ray appliance has limited means of determining if this happens.

More About the Sun Ray LAN Backnet

As mentioned previously, a Sun Ray backnet must be almost error-free, have sufficient capacity, low latency, and ensure that packets arrive in the same order as they were sent. In principle, this criteria rules out shared media (such as coaxial cables, hubs and fan-outs), even though they might be supported.

To ensure that the demands are met, you should use modern Ethernet switches with duplex capabilities and connect only one Sun Ray appliance per port. The sum of all traffic generated by the clients should not exceed the capacity of the network interface of the Sun Ray server (see"Sufficient Bandwidth" on page 6). Today, many switches use VLAN technology. So, for the continuity of this article, you must be aware of the functions and features of VLANs.

VLANs From a Network Point of View

A VLAN-capable switch can determine, based on certain configuration criteria, where traffic through it belongs. When the criteria are met, traffic is sent out through one or more ports. If the criteria are not met, the switch prevents the traffic from exiting the ports.

The three most common types of VLANs are:

  • Per-protocol based VLANs
  • Tagged VLANs (IEEE 801.2Q)
  • Per-port VLANs

In the first case, you can set up a switch to separate different types of networking protocols (such as TCP/IP, Appletalk, XMS and so forth). Because we deal solely with TCP/IP, this article does not discuss the other protocols.

In the tagging mode, the switch can inject (and discover) LAN identifiers in the header of an Ethernet packet, and route the packet to the correct location. The typical use for tagged VLANs is to be able to use an interswitch trunk to transfer several VLANs over the same line. Because the switch must open every packet going through a tagged port, latency could increase. However, high performance switches can easily handle this load, even at Gigabit wire speed. For large networks where Sun Ray appliances are just a part of the entire network traffic, and that span a large number of switches, the compound complexity could be a problem.

The last type of VLANs are the most important for this paper. Per-port VLANs can partition a switch into separate LANs. These LANs are logically separated, but physically belong to the same switch. With per-port VLANs, the physical and the logical architecture differ.

FIGURE 2 shows two LANs (A and B) using the same switch. Traffic does not flow from A to B inside the switch. It is separated as though A and B were two separate LANs. The example could represent a minor Sun Ray deployment where A could be the frontnet and B the backnet. A Sun Ray server should, in this case, have the frontnet interface connected to an A port (for example port 1) and the backnet interface connected to a B port (such as port 2).

Figure 2FIGURE 2 Physical and Logical Views of a VLAN Configuration on an Ethernet Switch

Long Distance Backnets

The distance between a Sun Ray appliance and its server is the sum of the segment lengths between them. For twisted pair, such a segment length can be approximately 100 meters. If the Sun Ray appliances must be at long distances, other media must be used, or there will be many needless interswitch trunk hops that increase latency. In cases such as these, fiber technologies are used. Two types of fibers are used for data communications, multimode fibers for ranges up to 2 km, and single-mode fibers for ranges up to (and, in some cases, well exceeding) 60 km. The fiber is not bound to run TCP/IP as such, but rather, it is a general media.

A fiber segment has no significant added latency, even when run over great distances, because it operates with laser generated light. Light in glass is sufficiently fast to enable us to disregard the latency generated by the distances for which they are designed. You can safely assume that the distance-related latency for 60 km is zero.

Summary of Sun Ray Networking

Great care must be taken when designing Sun Ray backnets to ensure that they have very few errors and low latency, deliver packets in the correct order, and have sufficient capacity. Per-port VLANs can be used to separate a Sun Ray backnet from other networks, and with care, tagged VLANs could be used to tie switches together within a building or compound. If you fulfill the design criteria, Sun Ray appliances can be deployed over longer distances using fiber technology.

  • + Share This
  • 🔖 Save To Your Account