HPE Community read-only access December 15, 2018
This is a maintenance upgrade. You will be able to read articles and posts, but not post or reply.
Hours:
Dec 15, 4:00 am to 10:00 am UTC
Dec 14, 10:00 pm CST to Dec 15, 4:00 am CST
Dec 14, 8:00 pm PST to Dec 15, 2:00 am PST
BladeSystem Virtual Connect
cancel
Showing results for 
Search instead for 
Did you mean: 

VC Flexfabric network latency measurement question.

 
chuckk281
Trusted Contributor

VC Flexfabric network latency measurement question.

Wang had a POC issue in measurung latency for interconnects:

 

*****************

 

We need help as we met a network latency issue in POC with VC FlexFabric , BL460Gen8 & BL660Gne8.

We find in VC docs that VC FlexFabric can be  “Low latency (1.5 μs Ethernet ports and 2.0 μs Enet/Fibre Channel ports) throughput provides switch-like performance”.

But in our testing environment , we got 170-200μs when we use ping between BL660 Gen8 and BL460 which are all in one C700 enclosure with VC FF.

The customer is futures exchange company and their application is sensitive with network latency. No matter how we changed the network config on VC, we still can not find any way to reduce the network latency . Is that the normal value (170-200μs) for VC FF’s Ethernet traffic?

 

Our testing environment: one C7000 with 2 x VC FlexFabric , One BL460c Gen8 & one BL660c Gen8 all with 554FLB, OS is Redhat 6.3.

 

**************

 

Some info for Wang:

 

**************

 

From Dave:

Are you using the ‘Private Network’ check box on that VLAN/Network(s)? 

That would force traffic to exit the enclosure to an external switch and then return.

This would cause some latency…

 

From Dan:

Ping will not give you switch latency.

 

Ping goes from OS1 through Network Driver to NIC1 to VC to NIC2 through network driver to OS2 and then all the way back again.

 

So before pointing the finger, you need to check driver stack as well and also look into OS latency optimization.

 

I would also verify you are not going across VC1 to VC2 during the ping (teaming involved?)

 

And from Alex:

Latency quoted in the QuickSpecs is a measurement for how long it takes for the data to enter a port and exit same port. This is not something that can be measure by a ping, since ping as an application at the OS level, as Dan said below.

 

You also have to consider payload size, etc.

 

******************

 

Other comments or suggestions?