Re: Ref tweaking

Brian Godette (bgodette@idcomm.com)
Thu, 19 Mar 1998 19:35:23 -0700


At 01:05 PM 3/19/98 -0500, you wrote:
>I'm running a reflector using Enhanced Reflector v1.07. Everything has
>been going fine but I'd like to know if there are things I can do to make
>sure the reflector is performing optimally, and I'd like to know how
>certain commands affect ref performance.
>
>First logging; does logging impact ref performance? If so, If I add the
>LOG-CONF-CONNECT command will that impact even further? Are there certain
>"LOG" commands that impact ref performance more so than others?

Not really, it does increase the CPU utilization a tiny bit as more lines
are being put in the log file, but that's about it.

>
>Can I use the "RATE-ADAPT" command to improve performance? I don't
>understand what this particular command does.

Using RATE-ADAPT is really recommended as the old Cornell defaults don't
reflect the reality of anything outside a LAN environment. However setting
meaningful values is what LOG-RATE-ADAPT is for, it prints out to the
logfile what the rate adaption code is doing, and shows things like loss
percentages, actual and reported bytes sent/received and rates.

The rule I've been using is to turn on log-rate-adapt, look at the
inbound/outbound loss rate at during various levels of activity to judge
what the no-loss-growth and loss-threshold should be. So far on most
reflectors I've had anything to do with adjusting, it's been RATE-ADAPT 10
20. Also note that ERef's behave differently in regards to the
no-loss-growth, it will increase the rate at which it sends if the loss is
*below* no-loss-growth, not simply when the loss is zero. This is so
intermittent periods of high loss doesn't floor a client's receive rate if
they have a small (1-2%) amount of sustained loss, which can and does
happen on WP and Cornell refs (tho MPCS seems to reset the xmit rate every
few minutes).