by Tim Greene
Juniper Networks has introduced its new data center architecture called QFabric, which promises improved performance and economy. Here are some questions and answers about the new technology.
What is QFabric?
It is Juniper’s new data center architecture that creates a single logical switch that connects the entire data center rather than tiers of multiple access aggregation and core switches.
Why is that an advantage?
It improves performance by reducing latency, which is accomplished by getting rid of layers of switches and cutting the number of devices needed, which also reduces the demand for space and power as well as management and maintenance.
What devices make up QFabric?
The architecture deconstructs a traditional switch into three component parts, an ingress-egress platform called a node, an interconnect platform corresponding to a switch backplane and a management platform that gives a single view of the fabric.
How does this flatten the architecture?
The node has switching and routing intelligence about all relevant paths through the fabric and makes Layer 2 and 3 forwarding decisions. Since nodes are logically connected peers, there is no need for aggregating ports with banks of access switches that feed aggregation switches that feed core switches. Packets are switched directly to the appropriate egress port.
Why is this fabric faster?
Packets are processed once and forwarded to the appropriate port without having to be reprocessed at every layer of a hierarchical architecture. Packets get through the infrastructure in a single hop. Juniper says delay is 5 microsec maximum across a data center at maximum cable length. Shorter lengths yield 3.71 microsec delay, which Juniper says is less than in a chassis-based Ethernet switch.
How does QFabric deal with congestion?
With its knowledge of other nodes, each node can sense congestion as it develops and adjust its forwarding rate accordingly. Juniper won’t say exactly how this mechanism works because it is trying to patent it, but it promises that the whole fabric is lossless even under congestion.
What is the practical result of this congestion control?
By dealing with congestion dynamically and quickly, QFabric gets rid of the need to over design the network to accommodate it. That boosts network utilization to near 100%, Juniper says.
What are some numbers on how well this fabric performs?
The average data center – defined by IDC as having 3,000 servers – using QFabric rather than a traditional three-tiered architecture is six times faster, requires 22%fewer devices overall and incurs half the operating expenses.
Does this have implications for cloud architecture?
Yes, but not immediately. Juniper says that it will announce later a means to incorporate WAN connections into QFabric, making it possible to link dispersed data centers to support cloud services.
What is QFabric?
It is Juniper’s new data center architecture that creates a single logical switch that connects the entire data center rather than tiers of multiple access aggregation and core switches.
Why is that an advantage?
It improves performance by reducing latency, which is accomplished by getting rid of layers of switches and cutting the number of devices needed, which also reduces the demand for space and power as well as management and maintenance.
What devices make up QFabric?
The architecture deconstructs a traditional switch into three component parts, an ingress-egress platform called a node, an interconnect platform corresponding to a switch backplane and a management platform that gives a single view of the fabric.
How does this flatten the architecture?
The node has switching and routing intelligence about all relevant paths through the fabric and makes Layer 2 and 3 forwarding decisions. Since nodes are logically connected peers, there is no need for aggregating ports with banks of access switches that feed aggregation switches that feed core switches. Packets are switched directly to the appropriate egress port.
Why is this fabric faster?
Packets are processed once and forwarded to the appropriate port without having to be reprocessed at every layer of a hierarchical architecture. Packets get through the infrastructure in a single hop. Juniper says delay is 5 microsec maximum across a data center at maximum cable length. Shorter lengths yield 3.71 microsec delay, which Juniper says is less than in a chassis-based Ethernet switch.
How does QFabric deal with congestion?
With its knowledge of other nodes, each node can sense congestion as it develops and adjust its forwarding rate accordingly. Juniper won’t say exactly how this mechanism works because it is trying to patent it, but it promises that the whole fabric is lossless even under congestion.
What is the practical result of this congestion control?
By dealing with congestion dynamically and quickly, QFabric gets rid of the need to over design the network to accommodate it. That boosts network utilization to near 100%, Juniper says.
What are some numbers on how well this fabric performs?
The average data center – defined by IDC as having 3,000 servers – using QFabric rather than a traditional three-tiered architecture is six times faster, requires 22%fewer devices overall and incurs half the operating expenses.
Does this have implications for cloud architecture?
Yes, but not immediately. Juniper says that it will announce later a means to incorporate WAN connections into QFabric, making it possible to link dispersed data centers to support cloud services.
No comments:
Post a Comment