Hello,

I've been reading over some papers on the Buffer Bloat site.  This one in 
particular has been quite informative: 
https://www.bufferbloat.net/projects/codel/wiki/Best_practices_for_benchmarking_Codel_and_FQ_Codel/#tuning-fq-codel.
  I have a private LAN in which several systems are connected via an enterprise 
class 40Gb switch. I'm tuning the settings for fq_codel, the QDisc chosen to 
eliminate many problems seen earlier.  This article mentions how fq_codel is 
well tuned for 1 and 10Gb speeds and makes mention of slower speeds.  Are there 
suggestions about how to tune the parameters for 40Gb?  I'm adjusting the 
parameters and the BQL values attempting to achieve as close to 40Gb speeds as 
possible.  The systems are CentOS 7.6 (kernel 3.10.0) and a Mellanox NIC 
(Connectx 4).  The network is configured for 9000 byte MTUs.  In lieu of this, 
I'm setting the quantum value to 9014.



Andrew Falanga
ENGINEER - SIG SYSTEM SOFTWARE
MSIG Equipment Engineering
Micron Technology, Inc.
Office: +1 (208) 368-4022  Fax: (208)368-2807

Email: 
[email protected]
_______________________________________________
Bloat mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/bloat

Reply via email to