I set this all up clean yesterday on a fresh windows install on an old laptop (parts for the real server are on order, this is just to test).
I have five cameras, three set to 4mpx, two at 8mpx, all at one frame per second to limit performance issues.
The good news is the server ran flawlessly, I tested periodic snapshots and got exactly what I expected (I push those off site). Well, after finding I needed to manually install the c libraries (really? the install doesn't check?).
The bad news is that the performance was awful, bad enough that the desktop client kept crashing -- it would come to the splash screen and disappear entirely (process was gone), no errors, nothing. At least I suspect performance issues, the server was running 50-60% but hit 100% as I tried to start the desktop, but lack of error indication means I really do not know. The one thread I found in the forums related to similar startup was unresolved.
The cameras are set as direct to disk. And I confirmed they are at 1 fps. I rebooted several times, checked and rechecked various settings.
By way of comparison running Luxriot the same exact camera feeds keep the cpu between 10-15% busy, that's with motion detection on (software, high accuracy giving up performance), recording BOTH primary and secondary (just because it defaulted that way). If I also display all cameras on an RDP session (mirroring what I would do in BI) it runs about 45-50%, but I cannot get BI to even reliably run in that configuration.
Xeoma was a bit higher but comparable as well (I had the cameras at 2fps with it without issues).
But I don't like either of them because they are flakey.
Am I missing something with BI and performance? I really thought I would avoid "flakey" given its reputation.
I plan to have much more substantial hardware -- though still I had planned to reuse an older server, so not THAT fast. I also planned to record at about 4 fps, so I am a long, long way from being able to sustain that.
With recording direct to disk, no transcoding, is performance that constraining? Or might I be missing something?
I have read though lots of discussions and almost all seem to conclude going direct to disk is most of the solution beyond the obvious of limiting resolution and frame rate.
The old laptop by the way has a new SSD, 8GB of memory (BI uses very little), I7 M620 @ 2.67ghz. Inbound camera feeds are about 8mbs total at the ethernet port (everything is wired, no wifi involved), disk activity shows only a few percent, so all the constraint appears on the CPU. Laptop was wiped first, so it has no competing processes other than vanilla windows 10 fully updated.