Choosing traffic shaping discipline for a small ISP

Hello

What is the best choice for traffic shaping when you have like 2500 users, on a single router with 50 Mbits/s internet fiber connection?
I've tried IPFW+dummynet, but browsing and download is not stable (variations are high), and second problem, with 98% cpu idle (Xeon CPU) with firewall activated, not all bandwidth is used, even on the evening when most of users are online.

I've tried pf + altq HFSC which works well for up to 500 users, for more users CPU is high, and browsing is very slow.
It is possible with CBQ to share internet for like 2500 users and limit their bandwidth? Both IPFW and PF when altq discipline needs sum of bandwidth for all users not to exceed total bandwidth (50 Mbps in my case) or well 100%, but having like 2500 and not 100 users, I cannot use values like 0.2%.

I did not posed rules, I've used standard shaping rules that can be found in any howto/man page, my question is only, how discipline should I use to get similar performances to HTB (or TC ) on Linux, for like 2000-2500 users.

Best Regards,
ovi

_______________________________________________
freebsd-isp@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-isp
To unsubscribe, send any mail to "freebsd-isp-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: C|Net on the Wintel "Bloatware" PC
    ... But in the dual CPU configuration, there's twice as many pin-outs to ... Intel 985X express bandwidth south ... This is why the architecture of the Motherboard design can't be so ...
    (comp.sys.mac.advocacy)
  • Re: TERPRI
    ... >> just worried about alienating some people over what is obviously a ... times slower than having the data and instructions all in CPU. ... bandwidth is much lower than RAM bandwidth which is slower than cache ...
    (comp.lang.lisp)
  • Re: DDR2 versus FBD
    ... George is it really your contention that the IBM tome dissing the ... Seemed like 2nd generation TOE ... My take is that TCP/IP offload removes work from the cpu - tada. ... bandwidth for more CPU wotk... ...
    (comp.sys.ibm.pc.hardware.chips)
  • Re: [Announce] [patch] Modular Scheduler Core and Completely Fair Scheduler [CFS]
    ... Implementations ... The first question that comes to my mind is whether nice levels should be linear or not. ... Maybe something like "each nice level gives X times the cpu of the previous"? ... It might be easier to specify the maximum deviation from the ideal bandwidth over a certain period. ...
    (Linux-Kernel)
  • Re: 16-Node Parallel System
    ... the interconnect of a SMP's 16 CPU interconnect will not ... >>match the bandwidth of the sum of the bandwidth of the 8 interconnects ... double core CPUs will cut it to 1/4. ...
    (comp.parallel.mpi)