so...what are these VCPU - what sort of horsepower are you dealing with? have you measured the actual ability of your REDIS sub-system - how many requests can it deal with directly? what CPU load do you get when doing that? have you profiled the radiusd process and seen what exactly is taking the CPU time when packets are going through your system? are things stuck in WAIT etc? alan On 18 January 2018 at 17:13, Alan DeKok <aland@deployingradius.com> wrote:
On Jan 18, 2018, at 11:55 AM, Matt Zagrabelny <mzagrabe@d.umn.edu> wrote:
For the sake of my own curiosity, does anyone know the throughput for FR dumping accounting packets to disk?
To the detail file? 1000's of packets per second. It doesn't do fsync, so it should as fast as memory / disk IO allows.
But, it all depends on the local configuration. OS, disk type, etc. It can vary widely.
We've done 10K+ writes per second to Redis. I don't imagine that plain disk IO is faster than that. So slower. but perhaps not by a factor of 10.
When you have millions of users and 10K packets/s, you need high availability, fail-over, clusters of servers, load balancers.
A "naive" install of "install FR and DB" is fine for 10K users, and 100 packets. It is entirely the wrong thing to do for 10K packets/s.
My final $0.02 is that it's not really nice to (a) charge a customer for services you're incapable of providing, and (b) ask *us* to support you for free, and then (c) complain about the free support.
Alan DeKok.
- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html