MetaFabric by Juniper

Screen Shot 2013-10-30 at 12.19.48 pm

Today Juniper Networks announced the MetaFabric. MetaFabric is the next step and evolution in data centre architectures. There has been a vast amount of chatter, noise and FUD recently when it comes to flow programmability and overlay networking. What Juniper has done is stayed clear of the Game of Thrones that Cisco and VMware have been playing which has allowed them to stay true to their vision.

MetaFabric is made of three components – the QFX Switch releases, Junos Space Network Director and recent upgrades to the Virtual Chassis Fabric. There is a raft of supplementary technology which gives the whole solution a little more shine also.

Screen Shot 2013-10-30 at 12.59.14 pm

The QFX5100 series updates are quite impressive. At first glance it is ‘just’ three new switches delivered into the Juniper DC portfolio but my oh my it is much more. All the QFX series push well over 1.4 Tbp/s throughput and clock shy of 1.1 Bpp/s and upwards. What is exciting for DC administrators, cloud architects and virtualization administrators is support for standards based networking protocols. virtualization protocols such as VXLAN, OVSDB, NVGRE are supported and VMware NSX, Contrail and OpenContrail are coming to the party. Q-fabric node integration is also supported bringing a bit of a well needed breath of fresh air to the Q-fabric ecosystem. Bringing such technologies to TOR plays well in favour of bridging the networking industries and virtualization industries technological apartheid.

Doug Hanks who wrote the Juniper MX Series for O’Reilly will be back with his next instalment I assumed entitled ‘Juniper QFX Series’ in June 2014. This is super exciting and pretty awesome – the calibre of the MX series book was phenomenal.

Screen Shot 2013-10-30 at 12.40.17 pm

Good old SPOG. It is an acronym I do not like and the notion of a single interface is a fallacy though that rant is saved for another time. Single Pane of Glass management irks me though it seems Juniper has tried to address a few of these concerns. Physical and virtual networks are documented and automation, restful API support and collection/correlation engines are supported. Junos Space Network Director aims to use API and automation techniques to allow integration and interaction with products like Puppet, Chef, Openstack, and VMWare. I think if this product grows well with feedback and is supported it may become the dominant pane of glass inside a Juniper DC – just not the only one!

Virtual Chassis Fabric is more than just Virtual Chassis. Someone inside Juniper has access to the same stuff Lance Armstrong was using. Virtual Chassis Fabric is the next step in Virtual Chassis evolution. Supporting Spine-leaf architectures  the Virtual Chassis Fabric architecture supports high performance low latency data centres with deterministic throughput and 2 microsecond latency. Leveraging a single point of management the Juniper DC can absorb older investments in the EX4300 and QFX3500/3600 series devices. With 2:1 port oversubscription the QFX5100 series provides great access layer capabilities. The 40GbE ports support breakout into four 10GbE ports that allow flexibility where required. This is great if yo currently need 10GbE and see migration to 40GbE within the product life cycle.

Screen Shot 2013-10-30 at 1.01.19 pm

The supplementary technologies that support MetaFabric such as TISSU has me really excited. Topology-Independent In-Service Software Upgrade (TISSU) allows the QFX5100 to run Junos within a VM on Linux. Junos is run on two separate VMs which default to active and standby pairs. This allows the device to ensure an intact data plane and continuous forwarding whilst the switches move to a newer software version. It is impressive to see this feature in TOR switching especially cutting reliance on other devices to accommodate traffic loads. Where I see TISSU being an additional benefit what is a feature I expect to see more of is the fact that Junos is being run as a Linux VM. This opens itself to a variety of new ideas. Can Contrail leverage this in a different way? Can other technologies be rethought with this shift?

It is a fresh take on a hotly contested market and the interest is keen. There was over 9000 requests for demo today alone for the QFX5100! Whilst many vendors sit in ivory towers sprouting the same stories Juniper has attempted to put into action their plan. I like where this is going and potential where Juniper is headed. It will be good to evaluate Juniper against Cisco and Avaya in the coming weeks but at this time it is a good step forward.

OpenContrail and Business intelligence

Networking Function virtualization (NFV) is what most businesses perceive as some of the strongest benefits of “SDN”. The ability to orchestrate and manage virtual network functions saves vast amounts of time, money, and can eliminate lag in delivery to market. NFV or service chaining allows the control of a sequence of events to happen to a flow.

For example a Cloud services provider might offer a Database workflow. This could be a small MySQL instance on Linux, a piece of storage, hidden behind a firewall, which requires access on port 443. The workflow would invoke the deployment of the Linux VM, attach and provision the storage, deploy a virtualised firewall with associated IP addresses, rules, and connections all off a single click.

The customers requirements aren’t just connectivity and who can access the database. They may be complex as If Customer A wants to access the database from internal, pass them through Firewall 1. If any customer wants to access the database from the internet, pass through load balance 1 or 2, firewall 1 or 2, and have the flow monitored by the IDS. Customers that access this database from the Gateway router are actually placed onto an overlay. Isolated, secure, and governed by the Contrail Control node.

The logical isolation of resources allows the multi-tenant data centre to enter a new era. Generally, all tenants use the same physical resources and hardware. These are servers, storage, and network. Now with Contrail and Openstack, all tenants are assigned their own logical resources. These logical resources are all isolated from each other unless specifically allowed. This is controlled by security policies

If you have started to visualise an architecture with Contrail the next step is exciting. You can look to start injecting business intelligence into your network. Think about some cases where you have attempted to meet a bunch of interesting constraints for a customer. Were they achievable or even repeated? Could it be automated and orchestrated into a repeatable workflow? The use of Contrail to take a vast subsection of smaller tasks, mix it in with business intelligence and orchestrate a workflow that will slash IT operation deployment times is of great appeal. The network has become smart once more.