8 Gbps FCOE and qos policy question

[gnijs]

Hello all,

When looking at the default qos policies applied on N5600, when you enable FCOE, FC traffic is getting 50% bandwidth reservation (with no drop policy) versus 50% default (ethernet) traffic.

But on a 10GE CNA, this would mean only 5 Gbps for FC, is this correct ?

Before going for FCOE, our servers were equipped with 8 Gbps FC HBA and some servers were using quiet a lot of this bandwidth.

If FC traffic is assigned "no drop" policy on a 10GE interface, how can you assure no drops if you only assign 50% of the bandwidth. If the server needs to send 6 Gbps FC traffic and 6 Gbps Ethernet traffic, the only way FCOE works is by giving FC the 6 Gbps and dropping Ethernet down to 4 Gbps. However, this is then in conflict with the policy (50% / 50% bandwith in case of congestion), hence my confusion....

[Walter Dey]

Your findings are absolutely correct.

However, you are aware of

- 8G FC means 6.8G at the datalink layer (encoding is 8B/10B)

- 10G Ethernet is 10G at the datalink layer (encoding is 64B/66B)

And most of the times, the bottleneck is the IOPS and not the bandwidth

[Paresh Gupta]

1. FCoE builds loss less fabric. That means it has mechanism (by PAUSE frames) to ask the server or storage array to back off when the network cannot accommodate more traffic. Hence, no drops would be observed on the fabric. In your cases, if storage disk starts pumping more traffic that 5G then the FCOE switch will send PAUSE frame asking it to slow down.

2. The configuration on switch is default. You can modify those. For example, you can set 80% BW to FCoE traffic. This is similar to the usual QOS configuration.

3. What Walter mentioned is also a factor here. At 10G speed, you get 50% more data throughput compared to 8G speed due to difference in encoding.