23 Feb 2011
Network vendor and Cisco rival Juniper Networks is aiming to take control of the datacentre switching market with a new network technology called quantum fabric (QFabric).
QFabric is Juniper Network’s new datacentre architecture. It is designed to offer a one-to-one connection to any datacentre element to deliver 10Gbit/s data transfer rates at less than five microsecond latencies, with no packet loss or jitter.
QFabric is the first fruit of Juniper Network's three-year Project Stratus research and development programme.
Juniper's fabric and switching group vice president Anjan Venkatramani said the programme "cost hundreds of millions of dollars to develop, consumed over a million man hours and led to 125 patents being filed".
The problem that Juniper Networks is aiming to solve with the Stratus technology is the tiered nature of datacentre networks.
The traditional datacentre architecture has core network switches, aggregation switches and top-of-rack (TOR) switches all layered in front of the servers delivering business applications.
This creates an environment that is hard to manage, particularly when it needs to be scaled up.
Juniper said that scaling up a traditional data centre is a case of “adding more tin”, without considering the adverse impact this has on manageability, power costs and data throughput.
The drivers for solving this problem are the increase in technologies such as cloud computing and the unquenchable demand for mobile internet services, especially data throughput, which require datacentre capacity.
Juniper’s QFabric has three components: a QF/Node which acts as the distributed decision engine of the network fabric, the QF/Interconnect high-speed transport device, and the QF/Director which delivers a common management window, controlling all devices as one.
Juniper launched the first product in the architecture today, the Juniper Networks QFX3500 – a top-of-rack system, representing the QF/Node part of Juniper’s architecture.
The QFX3500 is capable of operating as a standalone 64-port 10Git/s ethernet switch with fibre channel over ethernet (FCoE) and fibre channel gateway functionality.
The QF/Interconnect and QF/Director hardware systems will be available for order in the third quarter, Juniper said.
Quocirca’s principal communications analyst Rob Bamforth said: “Juniper's fabric approach is is a big shift, but it makes sense."
Juniper said QFabric architecture is up to ten times faster, uses 77 percent less power, requires 27 percent fewer networking devices, occupies 90 percent less datacentre floor space, and delivers a ninefold reduction in operating resources than the nearest competitive offering.
The competitive offering they cite is “is based upon product information that is publicly available for Cisco’s Nexus 7000 Series and Nexus 5000 Series platforms for a 6000 x 10 gigabit Ethernet port configuration standard industry estimate”.
It is no surprise that Juniper Networks is targeting market leader Cisco’s hold on datacentre switching infrastructure, but whether they can take Cisco's market share is another matter.
Commenting on the upcoming battle between Juniper and Cisco, Bamforth said: “This type of shift in thinking can undermine a major player, but not typically on its own, and might require the market leader to make slip-ups as well.
“Cisco is still looking pretty solid, but as other sectors have found, such as Nokia [in the mobile phone market], sometimes even apparently unassailable market leaders can stumble.”
This makes little to no sense to me.
a) Juniper introduces a switch based on merchant silicon (Broadcom) that they say they spent a million man-hours building. Since Stratus was announced in 2009 in February I guess we can assume that Juniper has had about 160 engineers working on this for 3 years - not really an impressive number of engineers. But I would hope that a merchant silicon based system takes less than that to turn up.
b) Am not sure how well this will scale with the small tables on the Broadcom switch chip, only 8000 host routes, and you need a host route for every IP in the subnet - i.e. the flat network. Virtualization uses a LOT of MAC/IPs... like 20-40 IPs per server... so this system is designed to support the ENTIRE DATACENTER on ONE 'SWITCH' with 200 servers? Hmm... something fishy.
c) the 40Gb is not even turned up yet, but isn't that what they are using for the proprietary uplink into the half-rack box in the aggregation tier? How is the network 'one switch' if you have multiple switches? Wouldn't this scale, in a 3:1 oversubscribed model then to roughly 500 ports (with table sizes for 200 in a resonably virtualized environment?)
d) So VMware's VP spoke, but I don't see the tie-ins to Virtualization? Is there auto vlan provisioning? Can we identify VMs from the network? What's the value?
Posted by: John McConnell 23 Feb 2011
Have your say on this article
Newsletters
Latest stories from Networks
Latest videos
You may also like
Networks jobs
Technology Patent Wars
Case studies from large organisations across all sectors
... And rich media, and flexible working, and peaks in traffic ...
Upcoming Events
Join us for this Computing web seminar, in which the Head of BI at the Co-operative Group Nick Colebourn will be explaining just how he reigned in the Group’s sprawling database estate and how significant savings were realised and data quality improved as a result.
Date: 31 May 2012
Time: 11:00 AM
Live June 13th 11:00am: Register now. During this web seminar we will be looking at the sorts of incidents that can bring data centres grinding to a halt and what can be done about them.
Date: 13 Jun 2012
Time: 11:00 am
Receive the latest jobs direct to your inbox
Are you being paid what you are worth?