I have been to a deeper presentation of the Brocade VDX switches that implement the VCS (Virtual Cluster Switching), and here is a resume of what that system does. This post is an updated and more complete than the previous post Brocade VCS or what I understood
VDX system consist of three concepts, Ethernet Fabric, Distributed Intelligence and Logical chassis
Brocade Ethernet Fabric has the following characteristic
- No spanning-tree protocol
- Multi-path deterministic
- Auto-healing, non disruptive
- Lossless and low latency
Distributed Intelligence has the following characteristic
- Self-forming arbitrary topology
- Fabric is aware of all members, devices and VMs
- Masterless control, no reconfiguration
Logical Chassis has the following characteristic
- Logically flattens and collapses networks layers
- Scale edge and manage as if one single switche (Coming in the next release)
- Auto-configuration (Coming in the next release)
- Centralized or distributed management
So after these marketing brief, here are my impression of the system.
- When 2 or more VDX switches are connected with each other they automatically form the Fabric, only configuration required is actually to define a VCS-ID (Default is 1) and R-Bridge ID (must be unique in the fabric) plus basic IP configuration for management.
- This is example 2 switches are connected with 3 10G links
- Note that each switch use WWN as an identifier, This come from brocade SAN switching business unit
- ISL Trunk (Note Cisco ISL trunk) are autoforming, no configuration are required, a maximum of 8 link can be in a trunk
- ISL trunks are packet based load balanced across the physical link in a round robin fashion, not based on 802.3ad algorithm. (I must trust the vendor as I could not sniff the traffic on the links)
- To form a ISL trunk, ports must be in the same ISL groups, witch are the following
- 24 ports
- Group 1 ports 1-12
- Group 2 port 13-24
- 60 ports
- Group 1 port 1-10
- Group 2 Port 11-20
- Group 3 Port 21-30
- Group 4 Port 31-40
- Group 5 Port 41-50
- Group 6 Port 51-60
- If you select ports that belong to different port group, 2 distinct ISL trunk will be form and the load sharing will be using the standard 802.3ad load balancing, typically using uneven link distribution
- Adding a new switch the fabrics, will add ISL link without any disruption of the traffic
Layer 2 routing
- VCS is based on IETF TRILL
- Brocade did not use ISIS for the routing but use SPF (Shortest path first), same use the brocade San products. When ask about IETF moving toward IS-IS as layer2 routing protocol, they stated that when standard will be ratified, they will comply.
- It is Hop based, therefore is you have 80G ISL going to 2 hops and a another 10G that is 1 Hop, the 10G will always be used
- Distributed Mac address table, In a VCS cluster, all switches shared the mac-address table
- Following is a show mac-address table from RB3. Note the destination port for mac 0050.563f.0001 and mac 0050.563f.0002
Automatic Migration of Port Profiles (AMPP)
- Port profiles are used to define the network configuration of a attached node (virtual or physical), it can include ACL, QOS policy, Vlan membership.
- Port profile will automatically move from one port to another ports in the switch fabric. This feature is usefull in a virtual environment where virtual server can be motioned from one physical box to an other
- In the figure above, the port profiles EsxDemo contain the following config
- Port is in access mode using Vlan1
- Access-list Esx_Demo_acl is applied
- Mac-address attached to the profiles are 0050.563f.0001, 0050.563f.0002, 0050.563f.0003, 0050.563f.0004, 0050.56a.6628 and 0050.567f.f669
- When connecting a non VCS switch to a VCS cluster, the Non-VCS switch see the VCS cluster as a unique switches, therefore link aggregation, Multi-chassis trunking can be used, This behaviour is the same with non Brocade equipment.
- Centralized Configuration, This is not available yes, Port profiles and configuration must be replicated to all switches forming the cluster. The centralized configuration will be in the next release od code (Q3 2011)
My comments and impression
- Since Trill is not fully ratified yet, Brocade do not use IS-IS for layer 2 mac address routing (TRILL), and I would have expected the routing to use the bandwidth instead of a basic Hop count, I have been impressed and please by the ISL and fabric auto-forming, Next release of code will support an automatic numbering for the R-bridge ID.
- The port profiles will simplify the configuration of the cluster as moving machine from one member to another require no configuration of the switch. Configuration only needs to be performed at the creation of the server and manually replicate to all cluster member. This will be a new twist for me and this might get tricky as I’m used to a more classic switch port configuration.
- The demo showed a 1 ping drop when moving the virtual server from 1 physical server to another, Brocade stated that they do not loose VOIP packet nor loose a video stream, Since the demonstration was over a Webex session, this was not shown
- Centralized configuration is a must to ease the configuration of port profiles as the VCS cluster can grow up to 12 switches and will eventually grow to 32 in later release. As for the Layer 3 function, this will be a nice to have but I see more these switches as top of rack datacenter environment
- In the road maps, are what is showed in the following slide, still need to be seen
- The product impressed me, but I can’t compare to any other manufacturer, as I did not see any demo from Cisco, Juniper and others. I found the technology interesting with a lot of potential especially if it is delivering what the marketing says. I will most likely see a deployment in the next months as one of our client already have a brocade cloud and he is considering these equipment for a new project
Reference and other blog about Brocade VSS