This post is more than 5 years old
1 Rookie
•
5.7K Posts
0
2356
How can I measure the max speed a dark fiber will eventuall support?
I have 2 Brocade DCX4S's connected over 1 ISL and I'm curious how much data per second I can send to the other site.
I used the commands spinfab and fcping to determine if the ISL is ok and what the latency is.
- spinfab returns that the ISL is fine
- fcping returns an average of 4.6 ms
The SFP's used are 4Gb long wave 4 km and we're using DWDM to span the distance.
The 2 sites are 42 and 62 km appart.
We are told that the active components of the DWDM will take care of the buffer credits and getting the data accross.
What do you think of this set up and what can we expect in the MB per second?
felipon1
96 Posts
0
July 13th, 2012 04:00
You're right. The switch calculates the distance in base to the echo delay measured in the ISL.
If you configure the ISL port as LD, with command portshow you'll see the stimated distance (with portbuffershow you'll see the BB credits reserved as well). Also the switch will provision BB credits for that distance, but it is a best-practice to manually calculate the real BB credit needs, because switch calculations can understimate the real bb credit needs.
felipon1
96 Posts
0
July 13th, 2012 01:00
Hi,
DWDM is a layer-1 technology,so to speak, and BB credits are transparent to it (as far as transporting them is concerned). If you've been told that the DWDM array supports 4G links, you should expect the same BW as if the remote switch were in the same Datacenter (with a 4.6 ms delay...). As you well know, configuring the correct number of BB credits will also help achieving a good performance.
Rgds
RRR
1 Rookie
1 Rookie
•
5.7K Posts
0
July 13th, 2012 03:00
Thank you, Felipon,
I'll go ahead and play with the buffer credits. I thought a switch could automatically detect the distance between the 2 sites, but somehow we don't get to see that. We were suspecting that we saw the distance between the switch and the DWDM devices.
RRR
1 Rookie
1 Rookie
•
5.7K Posts
0
July 13th, 2012 06:00
Thanks, I’ll do it manually. We did it automatically at first.
RRR
1 Rookie
1 Rookie
•
5.7K Posts
0
July 16th, 2012 04:00
I disabled the R_RDY and now the runtime of the spinfab is much shorter! Now a default spinfab runs 65 seconds instead of the 15 minutes or so which we saw before the weekend. Also I've increased the buffer credits to represent 120km. The counters in the GUI say only 550Mb is sent, but 10 million frames is 20GB, right? So why is only 550MB sent over the line?
RRR
1 Rookie
1 Rookie
•
5.7K Posts
0
July 16th, 2012 05:00
The 4.x ms response times remain, but the test is over a lot quicker, so disabling R_RDY seemed to have done the trick. Only we're seeing only less than 600MB being transported instead of the 20GB we were expecting. This represents about 64 bytes per packet. I'm going to try a spinfab using the full load of 2112 bytes per frame / packet. I'm curious how long it's going to take then.