What’s The Latency Really?

(Originally posted 2015-08-19.)

In What’s The Latency, Kenneth? I talked about Coupling Facility Link distance and OA37826. The whole supposition was that you might want to know about really long links.

A recent situation showed me that short distances might be a different and interesting matter.

So what’s a microsecond or two amongst friends?

Well, 1μs represents 100m of distance and 2μs represents 200m. This is, of course, as the fibre-bound photon flies. 🙂

But consider that RMF (and CMF for that matter) records latency in integer microseconds. And that the lowest value we record is 1μs meaning “no distance”.

I would hazard that “1μs” really means “from zero metres to 150 metres or so” and that “2μs” really means “ from 150 metres or so to 250 metres or so”. I’ve added the “or so” because I don’t think the 10μs per kilometre number is accurate to many significant figures.

But why does this matter?

In this situation the customer currently has a pair of zEC12 machines showing 1μs and 2μs latencies – depending on the path. All the links are Infiniband 1X HCA3-O LR (as in “long range”) links.

They’d like to move to z13 with perhaps Integrated Coupling Adapter ICA-O SR (“short range”) links or even 12X HCA3-O SR links. Both of these have maximum distances of 150 metres. Both are designed to have better performance than 1X HCA3-O LR links.

You can see the problem right there: Will these technologies do the job?

The latency information isn’t accurate enough to tell us. But nobody said it would be.

But hardware planners aren’t stupid either; They probably would’ve used 12X if they could.

Oh well, time to get the tape measure out 🙂 and see if we can get to under 150m.