There are certain limitations to pci passthrough, like no active snapshots, no vmotion. Those are slightly annoying to me but far outweighed by the other benefits.
I mean think about it. It doesn’t make sense to vmotion a server to other hardware if the card is installed on server A.
And that is why all the smart people use gateway devices like the Sangoma Vega 100.
I wanna be one of the smart people!
I’ve actually wanted to move our pbx over to a virtual machine for a while now, but was concerned about PCI passthrough problems, especially with some of our older servers. Are you saying that a Vega 100g would do and take the place of a Digium - Wildcard TE131/TE133 Card…which is what I’m running?
If you have already a TE131/TE133 card then it is really smart to use the dynamic routing (TDMoE) between the host and the VM (it works on any thing that uses “layer 2 networking” 
) and will likely save thousands of dollars of unnecessary hardware , it will also cost you not a penny (well a little bit of RTFM and then adding a few lines of code on your host and your VM’s DAHDI configs)
IMHO , I call this smart AND economical thinking, and anyone who has any dahdi hardware but has not yet bothered to try it is surely only showing half their smartness Why don’t you just “try it”
Hmm, so keep the existing freepbx going and setup a virtual pbx test box using TDMoE to connect to the TE131? Is that kinda what you’re saying?
Yes, you are simply decoupling the DAHDI hardware driver’s traffic between the Provider (the telco) and the consumer (asterisk) with a very small latency, Dahdi has had that in mind since its inception (Zaptel) if you have a solid network with an uninterrupted 2MB/s bandwidth, you are good to go for an E1, if you are using a Host/VM that would be an absolute no brainer , when it was first designed, then a 10Mbs ethernet was standard for a LAN ,so then it might peg out at two or three T1/E1 spans (each raw channel being of course 64kb) Nowadays, then NO PROB and your use case then the traffic can be over a random unused otherwise connection on a not-networked local ethernet channel !!
If of course you don’t have the hardware , then a TDM gateway is more versatile as it sits on the network not the hardware bus and decouples the TDM to SIP, it just costs more to do all that of course.
Just in case you don’t have TDM, the concept works equally for FXO/FXS traffic.
I appreciate the insight. I’m assuming that is how the Vega works…connecting to a freepbx box by TDMoE, just with a fancy freepbx module to help you set it up? And is something like https://www.voip-info.org/asterisk-tdmoe still relevant?
Now on the flip side, since this phone system is only a small part of my duties in the school district, I see that the VEGA 100 is about $1000-1200 and I have the server hardware. So, with that being said, I’m not sure which way to go.
No , I edited my post to clarify , the Vega box converts the TDM traffic to SIP, so there must be small time domain loss (likely not noticable though)
Your call, does 1000-1200 cover the cost of the learning curve ( small if you habe ever tried it) dahdi T1’s cost only a few dozen dollars on ebay/Amazon ( keep away from the Chinese clones ) so get a few for “belt and braces” solidity.
I would just “suck it and see” it would likely only take a Saturday afternoon to prove the concept.
No, because that ties the PBX to that piece of hardware. You lose the migration and fail over functionality that being a VM gets you. Heck you could lose the entire server room and spin up a backup in the cloud and just point a Vega100 to that and be back up and running.
Assuming the demarc was not in the server room and the telco hardware is still running.
The Vega is not a card that you physically plug into your server but a device that sits on your network.
You connect your PRI to it and configure a sip trunk to it from Freepbx.
I can’t agree that you need loose anything in a HA environment, it just use a different concept, I once used Sangoma hardware, and corosync, parallel two of the 8 port cards ($10000 in Vega terms, a couple thousand in Sangoma tdm pci terms) ) on your HA nodes and switch the active one to low impedance when active, seriously, it works and is really quite cheap (ProxMox, glusterfs over zfs or cepth all open source and you get to be in control of your own domain)
But if your solution works for you, then that is the glory of it, there is no “my way or the highway”, there is only "I go the High Road, you go the low road and the one who gets there last gets to buy dinner if they have any money left - "
An added advantage is the the first point of failure dos not include the hardware attached to the PRI’s as yours does’nt cover , making a dual Vega failover switch to the Telco is really not cheap or even available since the digium r400 ( whatever ) went away.
I appreciate all the info from everyone. It helps to look at it from different perspectives and get some opinions. I like the “cheap” approach idea, mainly because I am pretty cheap, but I do like some of the options brought up about the VEGA. Thanks again.
There are many “PRI gateways” available,
https://www.voipsupply.com/voip-gateways/digital
But none of them provide the basic requirements of “Highly Available” since the demise of the RedFone stuff. It’s just a fact IMHO, If your gateway dies, you are seriously “effed”, the only effective “HA” solution I found is the parallel Sangoma hardware in High impedance concept. Xorcom still do have an expensive TwinStar USB solution perhaps, but Seriously?
The reality is (at least in NANP land) is that each and every provider will bend over backwards to persuade you to accept their Cisco/whatever replacement router that spits out a so called “PRI”, because the cost of supporting a PRI over copper becomes increasingly expensive for them, and every competent TDM engineer is now retired or dead (except for Ralph ;-))
I suggest you hold a hard line, and insist they do so at THEIR cost, and at that point, the service will be internally delivered over SIP, it is then relatively easy to re-negotiate a direct SIP trunk to them ,at least that has been my experience.
Good luck with that , be strong and insist
So please explain why the posible demise of your Vega is covered in any way in your concept of ‘HA’ . . . And why that is ‘smart’
for anyone in any other way. . . .
You would have a secondary gateway. You need two PRI cards so having two gateways isnt that far of a stretch.
What happens when the PRI itself goes down?
The pri appears on two pairs of wires, there are only physical wires to connect them to hardware, gateways do not allow parellel connections, what are you missing here apart from basic mis-understanding of DC theory? High impedance allows the passive side to listen quietly, there could be many such parellel listeners, there can only be one talker to the pri without blowing things up, when the link goes down. The passive listener can see that, and cleverly realise such, it will then, with the connivance of corosync, spring into action. Change the interface to low impedance, thus pick up the span.
Missing the point. I didn’t assume they would both be on the PRI circuit at once. Having a secondary gateway onsite if something does happen to the first isn’t out of the realm. Does it mean there is a bit of down time while the connection is moved over? Yes but it still a solution.
But my question still went unanswered. What happens with the actual PRI circuit, itself, goes down?
It’s Dicko. He’s always contrary.
Then if you are clever, the d channel will report that even if you aren’t watching 
, your scripting can investigate, and if the alarm is far end , you will email/sms a sentient who will know to inform the provider. If local, you can panic as much as you want, it is just how PRI’s work
He certainly is, but he knows more than we do , or so he says