Leginon

Out of context, I can not tell which processing server you are talking about. It is certainly not myamiweb, that is the webserver.

Leginon documentation include a low-requirement processing server which is just a computer in the scope room with terminal to open the data acquisition wxpython gui for data acquisition. All images do pass through it, but the kind of processing it does is minimal in comparison to frame processing by Motioncor2.

The frame processing server is the high-requirement processing server and is likely what you refer to. It needs beefy GPU etc. It sounds to me that your setup put the whole leginon system, including myamiweb and database there. It is common to do this now since the rest of the requirement is small and a beefy GPU system can handle them at the same time.

The rules are:

All the GPU requirement is for Motioncor2.

The GPU requirement depends on how the data are acquired.

K2 super resolution mode needs 3-4 times GPU resource of counted mode for frame alignment.
1 focus + 4 image/beam shift data collection takes about the same time as 1 focus + 1 stage shift data collection but will need more GPU resource for frame alignment.

Typically, we find that we need 2 GPU for counted movie using 1 focus + 4 image/beam shift data collection scheme for a scope that runs 24-7.

Therefore 1 beefy 4-GPU system is likely enough for 2 microscope.

Webserver and database server, and data acquisition itself, if on the same computer, uses mostly cpu and cpu memory and is likely o.k. for a while even if you connect to two scopes.

What you may want to consider, though, is which gpu will the user use to display the gui during data collection if this is the only workstation. It is a waste to use high-end gpu for this purpose, especially now there will be two users at the same time.

Thanks, that was helpful. The 4-GPU system is a headless rackmount system in the server closet, so GUI display during data collection will likely need to occur elsewhere.

In that case,you may want to get a desktop computer that only has the Leginon low-requirement processing component (i.e., just the python part) installed, one for each microscope. We tend to use the older computers, low resolution monitor in the lab for this and send our money on the server side. Some labs buy a better one so that they can have dual usage.

So, now I'm a bit confused again. We will have a microscope PC and camera PC for each microscope/camera pair. Are you saying we need yet a third PC in the control room? There is already such a PC, used to interact with the microscope PC via TeamViewer, as per the vendors instructions. Can this be used as the third PC, assuming you mean a third PC?

The one I am referring to is usually a PC running linux. The microscopy support PC, the one the vender put it might work if they let you install a virtual box on it. I have never tried. Having a third PC is easier. No one except the service engineer needs access to the support PC if you are using Leginon.

See Network_Configuration

None of those network configurations make sense to me right now, most particularly the third one, where you essentially have 2 routes to the WAN. It would make more sense to use port forwarding or NAT on the (dotted line) linux machine, if this is being handled by a single machine. But back to my question. Using the Network Configuration as a point of reference, we are currently planning to have everything is the dotted line box run on a single linux machine. Your comment about about having the "leginon low-requirement processing component" run elsewhere is still confusing me and causing concern. I thought everything was managed through the myamiweb interface -- this is what my users have been telling me. It sounds like I'm missing something, and have at this point read through all the available documentation at least once. Simplified question: is there a step in the linux processing part which requires direct access to a GUI? This still doesn't mean a separate machine -- we could always run X2Go on our rackmount 4 GPU server, but I'm just more concerned that I still don't understand how all the pieces fit together.