Browsed by
Tag: FEX

Cisco UCS: Virtual Interface Cards & VM-FEX

Cisco UCS: Virtual Interface Cards & VM-FEX

Hello once again! Today I decided to talk about some Cisco innovations around of UCS platform. I’m going to try my best to keep this post high-level and EASY to understand as most things “virtual” can get fairly complex.

First up is Virtual Interface Card (VIC). This is Cisco’s answer to 1:1 mapped blade mezzanine cards in blade servers and other “virtual connectivity” mezzanine solutions. Instead of having a single MEZZ/NIC mapped to a specific internal/external switch/interconnect we developed a vNIC optimized for virtualized environments. At the heart of this technology is FCoE and 10GBASE-KR backplane Ethernet. In the case of the VIC 1240, we have 4x 10G connections that connect to the FEX, this connectivity is FCoE until the traffic gets to the fabric interconnect outside the chassis. The internal mapping to the server/blade allows you to dynamically create up to 128 PCIe virtual interfaces. Now here is the best part, you can define the interface type (NIC/HBA) and the identity (MAC/WWN). What does that mean? Easy policy based, stateless, and agile server provisioning. Does one really need 128 interfaces per server??? Perhaps in an ESX host you want the “flexibility and scale”. Oh yea, there is ANOTHER VIC that supports 256 vNICs and has 80Gbps to the backplane!!! That model is the 1280 VIC.

NOTE: 8 interfaces are reserved on both the 128/256 VICs for internal use and the actual number of vNICs presented to the server may be limited by the OS. 

Update: 

Just had a great conversation with a customer today and I want to take a minute to break down the math.

Today we have the 2208 FEX (I/O) module for the 5108 chassis. Each one supports 80G (8×10) uplinks to the Fabric Interconnect. This give a total of 160G to each chassis if all uplinks were utilized.

On the back side of each 2208 I/O is 32 10G ports (downlinks) for a total of 320G to the midplane. We are now at 640G total (A/B side). Take the total amount of blades per chassis and multiple that by 80G. 8 (blades) * 80G (eight traces per blade of 10G) = 640G. 🙂

Just keep in mind that the eight traces to each blades are 4x10G on the (A) side and 4x10G on the (B) side.

OK great I got all this bandwidth in the chasis, what can I do with all that? How about we carve out some vNICs. With the VIC 1240 mezz card you got 128 vNICs and 40Gb to the fabric. Not good enough? How about the VIC 1280 with 256 vNICs and 80Gb to the fabric. Just remember that your vNICs are going to have an active path mapped to either side (A/B) and can fail over to the other side in the event of an issue.  All the (A) side active side vNICs are in a hardware portchannel. Conversely the same holds true for the (B) side vNICs.

So Shaun, what’s you point to all this math? Choice and flexibility. You want 20Gb to the blade, you got it. You want 40G to the blade, done. 80G to the blade, no problem. 160G to the blade, OK but it has to be a full width. <GRIN>

Cisco: Nexus 2000 (FEX) Configurations

Cisco: Nexus 2000 (FEX) Configurations

It’s been way too long since I posted on my blog. Well, I have been studying for the CCIE Data Center lab and wanted to pass on some very critical information on fabric extender (FEX) configurations. One of the most common questions that our Cisco friends ask me is “Why do I need to create a port channel for the FEX-Fabric links?”. Well let’s dive into the WHY, and then explore the HOW.

Let’s first start with a foundation on what FEX is and how it works. FEX is a highly scalable, low latency data center access layer solution. What makes it so awesome? The fact that is managed as a line card vs. a separate ToR/EoR/MoR switch. Take your Nexus 7000 (core) or 5000 (agg/access) and they play the role of the “PARENT” switch. The FEX 2000 is essentially a remote line card to the parent switch. The FEX supports 1/10G and FCoE for consolidated I/O.

Nexus 2000 Comparison:

2248-TP (E) : This is the most common FEX, 48 ports of 100/1000 BASE-T host interfaces and 4 x 10G (SFP+) fabric uplinks. The (E) varient has additional shared buffer (32MB) locally vs. using what’s on the parent switch.

2224-TP: Same as above just 24 host ports instead of 48.

2148T: First generation FEX. The host interface can only operate at 1G. Not recommended any longer, go with the 2248-TP instead.

2232PP:  This is our de-facto 10G FEX. It has 32 x 1/10G and FCoE host interfaces (SFP+) and 8 fabric uplinks (SFP+). This switch also supports DCB.

2232TM: There is also a 10G BASE-T varient of the above FEX. THIS ONE DOES NOT SUPPORT FCoE, but does support DCB. 

OK, now that that is out of the way, let’s get back to the question at hand. How do I configure the connectivity to the parent switch?

I’ll get stright to the point. Use EtherChannel for the fabric uplink interfaces.

So, what are my options anyways?

Static Pinning and EtherChannel are your two options.

Why do we like EtherChannel fabric interfaces anyways?

Well the bottom line is that with Static Pinning you do not have automatic failover capability. Sure it’s deterministic, but if one of those fabric links goes down, so do the associated host interfaces. Let’s say your using two links and set the ‘pinning max-links 2’, half of the FEX host interfaces are mapped to one fabric uplink and the other half of the host ports is mapped to the other fabric uplink. Let’s look at a visual representation of static pinning.

The major issue with this configuration is that if the fabric uplink goes down, so do the ports associated to that interface. There is NO failover. This is WHY we want to use EtherChannel instead of static pinning.

Now let’s talk about the other (preferred) solution. EtherChannel is 110% the way to go. Let’s look at the visual representation.


As you can see from this diagram, we are load balancing the host interfaces (HIF) across all the fabric uplinks. Here is a note from the Cisco Configuration Guide.

Note

A fabric interface that fails in the EtherChannel will not trigger a change to the host interfaces. Traffic is automatically redistributed across the remaining links in the EtherChannel fabric interface.

When you configure the Fabric Extender to use an EtherChannel fabric interface connection to its parent switch, the switch load balances the traffic from the hosts that are connected to the host interface ports by using the following load-balancing criteria to select the link:

  • For a Layer 2 frame, the switch uses the source and destination MAC addresses.
  • For a Layer 3 frame, the switch uses the source and destination MAC addresses and the source and destination IP addresses.
    And finally the configuration (HOW) for all this awesomeness.

    1) ENABLE FEX GLOBALLY:
    N5K(config)# feature fex
    2) Configure the member interfaces for FEX connectivity:
    N5K(config)# interface e1/1, e1/2
    N5K(config-if)# switchport mode fex-fabric
    N5K(config-if)# fex associate 100
    N5K(config-if)# channel-group 100
    3) Create the EtherChannel Interface: This probably is done already based on the previous command, but it doesn’t hurt to make sure.
    N5K(config)# interface po100
    N5K(config-if)# switchport mode fex-fabric
    N5K(config-if)# fex associate 100
    4) Create the FEX:
    N5K(config)# fex 100
    N5K(config-fex)# description FEX_100<<PO100>>
    N5K(config-fex)# pinning max-links 1 (This must 1 for an EtherChannel configuration, if you change this to any other number you cannot use EC and your using static pinning instead)
    FOUR EASY STEPS!!!!
    Here are some good troubleshooting commands.
    sh fex”
    sh fex detail” – This command will show the HIF to Fabric Uplink Mappings and the version of code running on the FEX.
    sh interface fex-fabric”  – This command will display all the FEX units attached to the parent switch.
    sh inventory fex xxx” Display the inventory information about a specific FEX.
    show diagnostic result fex 100″ – Display diagnostic test results for a specific FEX.

NOTE: Nexus 7000 (Parent) FEX

If your trying to use FEX on the N7K be certain to issue the following commands or FEX WILL NOT BE ENABLED.

In the DEFAULT VDC issue this command “install feature-set fex
Now switch to the VDC that you want to enable FEX on and issue this command “feature-set fex

Your all set now.

Cisco.com Configuration Guide for Nexus 2000:
http://tinyurl.com/36uojrv