I need to reopen this discussion because it's impossible to run spandsp (and VoIP) under these circumstances. With zaptel unloaded, I see the following "vmstat 1" output: no swapping, an occasional disk output, +/- 1003 interrupts/sec., less than 10 context switches/sec., CPU idle 100%. A very quiet system. I load modules zaptel and wcfxo, and the system utilization stays the same. When I load wcfxs, the number of interrupts goes up to +-2004, which is normal. However, every three seconds the CPU spikes to 50%. This is "system" utilization, not userland. I assume it's in a wcfxs interrupt. The number of interrupts stays constant at about 2004 during each spike, leading me to the conclusion that the TDM card is holding an interrupt for 500ms every three seconds (50% of 1000ms is 500ms). This is a disaster for spandsp and VoIP in general. When I unload the wcfxs module, CPU idle goes back to a constant 100%. The TDM22B card is REV E/F, and I've tried it with several different cards. Fedora Core 3 with linux-2.6.9 downloaded from kernel.org (a stock kernel). The CPU is Athlon K7. Can anyone please give me a clue? Thanks, -- Michael Welter Introspect Telephony Corp. Denver, Colorado US +1.303.674.2575 mike@introspect.com www.introspect.com
If you are using an Athlon then you might have a VIA chipset and apparently non-intel chipsets can have these sorts of interrupt problems (Via especially). Try changing to an intel chipset motherboard. Craig ----- Original Message ----- From: "Michael Welter" <mike@introspect.com> To: "Asterisk Users Mailing List - Non-Commercial Discussion" <asterisk-users@lists.digium.com>; <support@digium.com> Sent: Tuesday, December 14, 2004 5:57 AM Subject: [Asterisk-Users] CPU spikes with wcfxs loaded> I need to reopen this discussion because it's impossible to run spandsp > (and VoIP) under these circumstances. > > With zaptel unloaded, I see the following "vmstat 1" output: > > no swapping, an occasional disk output, +/- 1003 interrupts/sec., less > than 10 context switches/sec., CPU idle 100%. A very quiet system. > > I load modules zaptel and wcfxo, and the system utilization stays the > same. When I load wcfxs, the number of interrupts goes up to +-2004, > which is normal. However, every three seconds the CPU spikes to 50%. > This is "system" utilization, not userland. I assume it's in a wcfxs > interrupt. > > The number of interrupts stays constant at about 2004 during each spike, > leading me to the conclusion that the TDM card is holding an interrupt > for 500ms every three seconds (50% of 1000ms is 500ms). This is a > disaster for spandsp and VoIP in general. > > When I unload the wcfxs module, CPU idle goes back to a constant 100%. > > The TDM22B card is REV E/F, and I've tried it with several different > cards. Fedora Core 3 with linux-2.6.9 downloaded from kernel.org (a > stock kernel). The CPU is Athlon K7. > > Can anyone please give me a clue? > > Thanks, > > -- > Michael Welter > Introspect Telephony Corp. > Denver, Colorado US > +1.303.674.2575 > mike@introspect.com > www.introspect.com > _______________________________________________ > Asterisk-Users mailing list > Asterisk-Users@lists.digium.com > http://lists.digium.com/mailman/listinfo/asterisk-users > To UNSUBSCRIBE or update options visit: > http://lists.digium.com/mailman/listinfo/asterisk-users >
> I need to reopen this discussion because it's impossible to run spandsp > (and VoIP) under these circumstances. > > With zaptel unloaded, I see the following "vmstat 1" output: > > no swapping, an occasional disk output, +/- 1003 interrupts/sec., less > than 10 context switches/sec., CPU idle 100%. A very quiet system. > > I load modules zaptel and wcfxo, and the system utilization stays the > same. When I load wcfxs, the number of interrupts goes up to +-2004, > which is normal. However, every three seconds the CPU spikes to 50%. > This is "system" utilization, not userland. I assume it's in a wcfxs > interrupt. > > The number of interrupts stays constant at about 2004 during each spike, > leading me to the conclusion that the TDM card is holding an interrupt > for 500ms every three seconds (50% of 1000ms is 500ms). This is a > disaster for spandsp and VoIP in general. > > When I unload the wcfxs module, CPU idle goes back to a constant 100%. > > The TDM22B card is REV E/F, and I've tried it with several different > cards. Fedora Core 3 with linux-2.6.9 downloaded from kernel.org (a > stock kernel). The CPU is Athlon K7. > > Can anyone please give me a clue?I see about the same thing with RHv9, single tdm04b (no fxs) rev e/f, 2.2ghz celery, and zero calls being processed. The spikes to not correspond with anything observable with a packet sniffer.
Craig Guy wrote:> If you are using an Athlon then you might have a VIA chipset and apparently > non-intel chipsets can have these sorts of interrupt problems (Via > especially). Try changing to an intel chipset motherboard. >On an Intel P4 board I still get the 3s spikes. The spikes are "wider" on the plot, i.e., taking more than a 1s interval. With irq balancing the interrupts alternate between processes. -- Michael Welter Introspect Telephony Corp. Denver, Colorado US +1.303.674.2575 mike@introspect.com www.introspect.com