vicidial.org

I have two asterisk servers that I've been testing various combinations of asterisk and vicidial with and this card pisses me off to no end. I know that there are several other more expensive cards I can choose from, but I am not sure if i will have the same annoyances with those.

I've gotten it working on both centos and opensuse, and now when i try on vicidial now, I can't get the damned card to configure properly anymore (it's insisting on sharing an IRQ all of a sudden). And even when it does work on the live machines, I've had two occasions where meetme seemed to die because the card somehow lost its irq (i had to change it back and forth between two slots to fix the problem), however asterisk stayed up and the people that weren't using the dialer were fine.

If I were to spend 750 on one of those digium pcie cards would i still have all these IRQ headaches? It's a little annoying to even be contemplating this since I don't use the card for anything but the timing. Otherwise does something like a Digium TDM400P work a lot better?

I'll throw my two cents in here... I've had the IRQ sharing problem with the X100P, the 1 port T1 cards, the 2 port T1 cards, and the 4 port T1 cards although I never have trouble configuring the cards. The only problems I run into are when the card is sharing an IRQ with a busy device it tends to drop calls and sometimes resets the circuits. At one point I purchased a handful of cool 1U servers that had the PCI slot hard coded to share an IRQ with the network card, these servers were very problem prone. If you have more than one server you might want to look into TDMoE, I've been using it without trouble on the same problematic servers mentioned above and it seems to work like a champ! If you only have one server, I would recommend turning off all unused devices in the BIOS (i.e. USB, serial, parallel, sound) this will free up some IRQs and then you can move the card between a few slots until you get good results. You can test with a shell application call "zttest". Let me know if you need any more help. One last note, I've had much better luck with the PCI-Express cards...

You can also check which device it's sharing IRQs with by running this command: "cat /proc/interrupts" of course without the quote signs.

mxtreme311 wrote:I'll throw my two cents in here... I've had the IRQ sharing problem with the X100P, the 1 port T1 cards, the 2 port T1 cards, and the 4 port T1 cards although I never have trouble configuring the cards. The only problems I run into are when the card is sharing an IRQ with a busy device it tends to drop calls and sometimes resets the circuits. At one point I purchased a handful of cool 1U servers that had the PCI slot hard coded to share an IRQ with the network card, these servers were very problem prone. If you have more than one server you might want to look into TDMoE, I've been using it without trouble on the same problematic servers mentioned above and it seems to work like a champ! If you only have one server, I would recommend turning off all unused devices in the BIOS (i.e. USB, serial, parallel, sound) this will free up some IRQs and then you can move the card between a few slots until you get good results. You can test with a shell application call "zttest". Let me know if you need any more help. One last note, I've had much better luck with the PCI-Express cards...

I've actually already done all of that and it's sitting on its own IRQ (but even that didn't work for vicidialnow; it insisted on sharing for some reason). It's been up for 14 days now with ~17 agents, which is the target I was aiming for before we'd move our inbound over to it, so i'm not really messing around with alternate configurations anymore.

As you said though, my last remaining problem seems to be randomly disconnected calls, which a bit of reading seems to tell me is related to "interrupt problems." I'm going to attempt to quantify it next week, but if I had to guess based on user comments, it's probably sitting at 0.25-0.5% of all calls, or 15-30 calls per day. It's insignificant for outbound, but I'd like to eliminate it for inbound as I place higher value on our customer service calls.

Are there any "low-end" pcie cards? My calls all go through SIP trunks so it's a bit of overkill to get one of the $750 cards... Pcie also makes sense because a lot of new server boards don't have pci slots, and I figure i'll be upgrading the machines at some point to more redundant hardware.

In what specific way were things better with pci-express cards? Can you give me manufacturer/models you've used with success? Before this, the last time I really had to concern myself with IRQs was in the early 90s on my 486.

We have used the Sangoma PCIe cards before for PCI express timing and they run about $300 for the lowest-cost version with one FXS module.

This really isn't a vicidial problem, but this may help others that are experiencing these issues. I've been running some tests and it seems my problems are related to my NIC. Even though the x100p is on its own IRQ, when i ran a 200MB file transfer while running zttest, I watched accuracy go to shit for the duration of the transfer. We're talking a continuous string of <20% accuracy for the duration. I'm gonna try disabling onboard ethernet and running zttest to see if that completely eliminates IRQ misses. If it does, I'll test with a pcie network card and see if there is still no problem. If I still have occasional misses I Guess i'll have no choice but to go with a pcie asterisk card.

edit: well, I swapped in an intel pro 1000 pcie card and have have zttest running while i ping -s 65505 -i 0.00001 my server with a few sip channels in calls and there hasn't bing a single blip on my zttest so far. It's been running for a good 45 minutes but I'm gonna leave it for an hour). Bottom line is I think my problems are solved so I'm pretty ecstatic and there's no need to purchase a $300+ Sangoma pcie card. I can't match our live call volume while testing so I guess I won't be 100% sure it's fixed until people go back to work next week, but I'm optimistic. ^_^

we have had much fun with x100 cards from x100p No matter what we did/where we put them they would nearly always share IRQ's with something, often fail to initialise properly at boot and the zttest scores were not particularly steady.

we currently have 1 dialer with an x100
4 with ztdummy (1000hz kernel timer) and 1 with a Digium wildcard TE212p (using the zap lines).

The one with the x100 is the only one that seems to suffer from "garbled" conferences once in a while. The others are all fine. FOr the hassle and no apparent improvement in sound quality I don't really see any use for the x100's

HTH

Chris