WHY IS QoS NEEDED?
WHY IS QoS NEEDED?
The primary goal of QoS is to provide priority for traffic flows to and from specific devices. In this context, priority means providing lower latency and higher bandwidth connections with more controlled jitter.
An underlying principle of Fibre Channel switching is that the network guarantees that no frames will be dropped. If this is the case, why do we need QoS at all?
Switches today provide high-performance, non-blocking, non-oversubscribed crossbar switch fabrics. The Cisco MDS 9513 Multilayer Director can switch more than a billion frames per second. Why would users ever need QoS when a switch fabric provides seemingly endless amounts of frame-switching capacity?
The answer is simple: congestion.
Congestion occurs for two basic reasons:
• Congestion will occur if multiple senders are contending with a smaller number of receivers. If the aggregate rate of traffic transmitted by senders exceeds the size of the connection to the receivers, blocking will occur (Figure 1).
• Any time there is a speed mismatch between senders and receivers, buffering will occur. Buffers are a finite resource on switches, typically in the range of 16 buffers (32 KB) to 255 buffers (512 KB) per port. When these buffers are full, blocking occurs (Figure 2).
Figure 1. Congestion Caused by Senders Outnumbering Receivers
Figure 2. Congestion Caused by Speed Mismatch between Senders & Receivers
Many organizations consolidate their SAN infrastructure in order to realize cost savings and increased management efficiencies by pooling disparate storage resources into one single physical storage fabric. Managing contention for resources is an important aspect in realizing the business benefits associated with storage consolidation. If storage resources become congested because noncritical business applications cause time-sensitive mission-critical applications to become slowed down, the cost benefits associated with SAN consolidation quickly disappear.
There is no automatic quick fix to alleviate congestion and blocking. It is possible to add more buffering to a switch, but additional buffers will not remove the congestion; they will simply increase the time it takes for congestion to turn into blocking. Virtual output queuing (VOQ) can be used to prevent one blocked receiver from affecting traffic being sent to other noncongested receivers ("head-of-line blocking"), but it does not do anything for traffic being sent to the congested device.
SAN traffic can be categorized as a large number of devices (hosts) communicating with a smaller number of devices (storage ports). Put another way, SAN designs are almost always over-subscribed, with more host-attached ports than storage-attached ports. What is important is making sure that the oversubscription does not impact the performance of mission-critical time-sensitive applications.
The primary goal of QoS is to provide priority for traffic flows to and from specific devices. In this context, priority means providing lower latency and higher bandwidth connections with more controlled jitter.
An underlying principle of Fibre Channel switching is that the network guarantees that no frames will be dropped. If this is the case, why do we need QoS at all?
Switches today provide high-performance, non-blocking, non-oversubscribed crossbar switch fabrics. The Cisco MDS 9513 Multilayer Director can switch more than a billion frames per second. Why would users ever need QoS when a switch fabric provides seemingly endless amounts of frame-switching capacity?
The answer is simple: congestion.
Congestion occurs for two basic reasons:
• Congestion will occur if multiple senders are contending with a smaller number of receivers. If the aggregate rate of traffic transmitted by senders exceeds the size of the connection to the receivers, blocking will occur (Figure 1).
• Any time there is a speed mismatch between senders and receivers, buffering will occur. Buffers are a finite resource on switches, typically in the range of 16 buffers (32 KB) to 255 buffers (512 KB) per port. When these buffers are full, blocking occurs (Figure 2).
Figure 1. Congestion Caused by Senders Outnumbering Receivers
Figure 2. Congestion Caused by Speed Mismatch between Senders & Receivers
Many organizations consolidate their SAN infrastructure in order to realize cost savings and increased management efficiencies by pooling disparate storage resources into one single physical storage fabric. Managing contention for resources is an important aspect in realizing the business benefits associated with storage consolidation. If storage resources become congested because noncritical business applications cause time-sensitive mission-critical applications to become slowed down, the cost benefits associated with SAN consolidation quickly disappear.
There is no automatic quick fix to alleviate congestion and blocking. It is possible to add more buffering to a switch, but additional buffers will not remove the congestion; they will simply increase the time it takes for congestion to turn into blocking. Virtual output queuing (VOQ) can be used to prevent one blocked receiver from affecting traffic being sent to other noncongested receivers ("head-of-line blocking"), but it does not do anything for traffic being sent to the congested device.
SAN traffic can be categorized as a large number of devices (hosts) communicating with a smaller number of devices (storage ports). Put another way, SAN designs are almost always over-subscribed, with more host-attached ports than storage-attached ports. What is important is making sure that the oversubscription does not impact the performance of mission-critical time-sensitive applications.
Comments