Newbie's V-Drums -- MIDI struggle

V-drum latency is a huge can of worms from my perspective. The above post is right. If you can monitor from the brain directly, that will always be the fastest. How you hear that with the song is another can of worm. So lets move the bar and say you want to monitor the Soft synth. Now things get more complex and laggy. Lets look at every step of the chain:
1. Trigger sensing time: I have found that 3.2 to 2.2 is the pocket with a Roland brain and trigger. Longer does not help, and shorter than 2.2 results in wild velocity results.
2. Most current brains offer both USB and 5 pin MIDI outputs but no specs are offered on either. good luck getting that info from Roland. I'm in process of switching from a Roland TD-30 to a Megadrum for that purpose, but I still have to test USB vs 5 pin because of the next step of the chain.
3. Into the computer: If your USB is used for many things, then USB gets bogged down. an Edirol UM-3EX is one fo the more stable 5 pin to USB converters if you want to try that rout. I just ordered an Alysium yesterday that uses copperlan to get 5 pin going into the computer using RJ-45.
4. Computer to DAW. The buffers in sonar is another lag. I'm running sonar at 64 buffers to minimize this, but I have a very strong Audio PC, most are running at 128 or 256 buffers. The lag is Buffers/sample rate. So the higher sample rate you run at will help this if your computer can go fast enough.
5. Processing in the VST. The more you ask the softsynth to do, the more lag it can generate.
6. Now you need to get the Audio out of the computer. Most converters have a fixed number of samples for the input and output. Again this is divided by the sample rate.
 
Assuming you don't have much going on with USB, the trigger time may or may not be the largest part of the lag. The largest lag and jitter component could be USB. Anything over 4.5 ms starts to feel bad. With 3.2 of that taken up by the triggers, it is very difficult and expensive to make the rest of the chain only take up 1.3 ms. 
 
I have been working on this for 5 years spending tons of cash. I elected for SSL alpha link coverters that only need 23 samples out (Only the QS RME line is faster at 8 samples ($$$$$$)). I also went with Megadrum trigger inputs, and an Alysium 5 pin to Network MIDI input. All to try and reduce that lag. I should be able to test it all in about a month. I also decided to work at 96K audio because past that takes up too much hard drive space and CPU overhead. 
 
Another detail to look at is the PPQ (Midi resolution) of the midi recording. Maxing it out at 960 may not be best because that puts to much strain on the CPU. Most go for 240 or 480 for best results depending on the tempo you are working at. It's all a balancing act and money pit.
 
Before you get to any of that, it also takes tons of time to dial in the velocities right. Triggers in general are not good a loud vs really loud. Once you dial in the triggers to  never overload, then you have hot spots to deal with that blow the velocity dial in out of the water. yet another balancing act that can be difficult to live with. I like BFD because of the velocity controls. In general, make the drums/triggers inside the brain sound good first before you ever try to dial in the VST. The VST velocity mapping gets used up trying to make the sounds work right.
 
I have yet to get to the Sonar routing of this. Should I have one track for MIDI only, rout the MIDI to the VST, then have another track that gets the Audio and then send that out? What if I want to blend sound on the drum brain with the VST? What if I want to use a real out board compressor on the snare like real drums? it all gets very complex.
 
I have plans to get my Edrum and real drums working side by side, then eventually trigger up my mic'd drums to get the best of both worlds.
 
My current setup started as an Edrum kit and has evolved into a full blown studio. Currently waiting on two 96 point patch bays, the Network MIDI adapter, and some cables that were all ordered over the last week.
 
Another trick I have been working on is rim splitting. Roland switch triggers only use one trigger and a switch to decide if the rim has been hit. This worked with a TD-20, but not on anything newer (I was bummed when I upgrade to the TD-30 to gain USB). You can use an insert cable and headphone splitter to ends the entire signal to one trigger in, and then only send the head signal without rim to another input. Then turn the head sound off for the main input, using the secondary input for the head sound. This allows you to dial in the trigger inputs for rim and head independently. Only the Roland brains understand how to interpret the rim right as there is no MIDI protocol for some of the things Roland brains do internally. I plan to use my TD-30 for rim sound only, and have the rest come from a Megadrum and BFD3. 
 
Hats are also a can of worms as stated above. I'm also thinking about a KAT DITI to get good hat controls.
 
Notes: I'm using Hart drums with the cones swapped out to Roalnd to re-gain position sensing. I also use a KD-140 with the head switched out to Magnums for less in room noise.