nanog mailing list archives

Re: Latency question - Juniper MX960 vs the Tellabs 8860


From: Steve Danello <the76posse () gmail com>
Date: Wed, 13 Feb 2019 13:32:51 -0500

Yes thank you Saku!   We’ve accounted for the fiber distance.   We just want to be able to quote the hardware latency.  
Thanks for verifying that Juniper; just need the Tellabs...  thank you:)

Sent from my iPhone

On Feb 13, 2019, at 12:53 PM, Saku Ytti <saku () ytti fi> wrote:

Hey Steve,

15us would be more than you'll realistically see in uncongested MX960,
but it is in the ball-bark. I've not tried Tellabs, if customer is
latency sensitive you probably want to look constant time pipeline
devices rather than run to completion npus. You'll see low single
digit us on typical box. Remember that 10us is just 2km of fibre (1km
of RTT).


On Wed, Feb 13, 2019 at 7:28 PM Steve Danello <the76posse () gmail com> wrote:

Looking for two latency profiles, best/worst cast would be great for the Telllabs 8860 and Juniper MX960.   Looks 
like the Juniper may be in the 15us range

Both cases would be a 10Gb -1Gb and 1Gb - 10Gb

We’d understand the serialization piece; really looking to provide a customer the hardware latency, we don’t have a 
great means to test them

Thanks all!

Steve



-- 
 ++ytti


Current thread: