Cisco UCS and I/O?

[ryabutler]

Hi,

I have a very quick question in regards to I/O discussed with Cisco UCS.

I often see I/O referrenced with servers with the number of operations it can perform.  Not with network performance or throughput.

But it seems like I/O discussed with Cisco UCS is talking LAN/SAN throughput for the I/O.

Is this the case?

Any help would be great for me to understand this better.

Thank you!

-rya

[mipetrin]

Hi,

Yes you are correct. With the UCS when it mentions, for example, 20 Gbps of I/O per server, this is in relation to network throughput. If you are using 2 x Fabric Interconnects, each server will have a 10 Gbps connection through Fabric A and another 10 Gbps connection through Fabric B. Thus providing up to 20 Gbps per server.

Hope that helps clarify things for you.

Regards,

Michael

[alan.ambers]

Michael - Let me try to understand this in our environment.

We have to UCS 5108 each fully populated with 8 1/2 blade hosts.  We also have two Fabric Interconnects.  Each Fabric Interconnect has two 10Gig connections to each FEX (IOM).  The "A" FI has it's two connection to the left (as viewed from the back) IOM on UCS "A".  FI "B" has two 10Gig connections to the right IOM on UCS"B".  This is for a total of 40Gig.  For reference, each FI has a 10Gig connection to our Cores for a total of 20Gig.  We will be soon doubling this.

My question is how the IOMs talk to the 8 physical servers.  If FI "A" goes down, then is there any lack of connectivity to whatever is connected to the left IOM?

Thanks!

/alan

[Robert Burns]

Alan,

Depending on which Mezz card and configuration you're using will dictact the behavior seen if you have an FI fail. Certain Mezz cards support "failover" config.  The failover config will allow UCS to "redirect" backplane traffic from a NIC from one IOM, to the other in the event it fails.  The host OS would NOT see a link failure in this regards.  Without the "failover" enabled, if an FI failed, and associated vNICs to that fabric would appear down to the OS.

Mezz cards that support failover are:

M71KR-E (or -Q) - First generation Emulex & QLogic CNA's

M81KR - Virtual Interface Card "Palo".

Mezz cards that do NOT support failover are:

M72KR-E (or Q) - Second generation Emulex & QLogic CNA's

Intel 10G

Broadcom

regards,

Robert

[KRIS PATE]

Is this the fabric failover that you are referring to?

Is it still the case that this isn't supported in a Vmware environment?

Kris

[mipetrin]

Hi Kris,

Yes, Rob was referring to Fabric Failover.

In regards to your second question, it is still recommended to run the Nexus 1000v within a UCS system without any adapter fabric failover. For a better understanding as to why this is, please review the following document:

Best Practices in Deploying Cisco Nexus 1000V Series Switches on Cisco UCS B Series Blade Servers

http://www.cisco.com/en/US/prod/collateral/switches/ps9441/ps9902/white_paper_c11-558242.html

Thanks,

Michael