FOCUS CLOUD NETWORKS
THE CLOUD NETWORK DEBATE
In February Juniper Networks announced QFabric. The company has positioned it as the Cloud-ready network architecture. FOCUS gathered reactions to the launch from Prof. Ashwin Gumaste. We then put his points to Anjan Venkatramani.
Ashwin Gumaste: The solution put forth by Juniper is only a comparative one against Cisco. But then there are others out there who are more proficient in this data center space and can lead to carrier-class performance at even lower latencies and better energy needs. The concepts of a data center being a monster switch is somewhat hard to digest, given that the data center will always scale and no single fabric can meet the scalability requirements. The larger concern I have is about being able to meet carrier-class performance within the data center.
Ashwin Gumaste: I don’t believe that this technology
solution is radical at all! In fact,
it is evolutionary - one step ahead of what exists in the market. To make progress we all take these baby steps, so this isnt something that makes the solution radical! What makes this interesting is Juniper’s
Issue 15, April/May
Anjan Venkatramani: Juniper’s QFabric is indeed a radically different approach to data center switching and is going to be very disruptive. While it presents itself as a single switch to the DC operator and to orchestration applications, it has a distributed implementation of both the data plane and control plane. The distributed implementation is very much like a network, the properties and abstraction are very much like a single switch. The QFabric single- tier architecture will have lower latencies and lower power than multi-tier implementations because there are just fewer processing stages.
desperation to show this as better than its competitors when it actually compares it to only one competitor and not everyone else!
Anjan Venkatramani: The comparisons and superiority of QFabric and a single-tier architecture are equally applicable to all vendors and all multi-tier architectures.
Anjan Venkatramani, VP Fabric and Switching Technologies at Juniper Networks
Laboratory, Indian Institute of Technology
Ashwin Gumaste: The other problem I see with switches of this large port density is that they present a single point of failure. Basically, if the switch goes down, thats the end of the data center - even if you add 1+1 redundancy, then the cost goes up and seamless movement is still difficult. Overall, this approach will create issues from a fault tolerance perspective. If you see cloud networks, they are all distributed. The very idea of doing a distributed network is to be able to keep smaller entities are discrete locations and be able to virtualize your entire infrastructure. This approach goes against this basic tenet of cloud computing.
Computer Science and Engineering, Convener, Gigabit Networking
Prof. Ashwin Gumaste, Ph.D. James R. Isaac Chair,Department of
Ashwin Gumaste: One would be more interested to know if this solution scales, and when it does what the performance guarantees are. The solution talks of 10 microseconds latency at full scale - I believe there are others out there who have a better latency - and perhaps scale significantly. Again, challenging Cisco is fine, but I think this should be more
about technology than one-upmanship, which is really lacking here.
Anjan Venkatramani: Juniper is not aware of any vendor which has a latency of 3.7us across 6K+ 10GE ports or less than 10us across tens of thousands of ports. Congestion management is built into the fabric.
Anjan Venkatramani: To reiterate, QFabric is a distributed implementation of a single switch. There is no single point of failure anywhere. The cost does not go up to implement redundancy. Seamless mobility is possible within and across multiple QFabrics. Also, the beauty of the design is that the choice is given to the customer to make QFabric as big as they want it to be – they can implement a 3-tier architecture, a 2-tier architecture or a 1-tier architecture with QFabric.
JUNIPER’S QFABRIC CLAIMS
1. Any-to-any connectivity with fairness and full non- blocking
2. 4.
If you want to join the discussion visit
www.datacenterdynamics.com/mydcd
34
www.datacenterdynamics.com 5. Low latency and jitter
3. No packet drops under congestion
Linear cost and power scaling
Support of virtual networks and services
6. Modular distributed implementation that is highly reliable and scalable
7. Single logical device
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64