Algorithms sit in Que for a long time and take a long time to process

If you check the Result dropdown and the queue is reporting zero(it usually is) then I suggest deleting and relaunching a backtest if it seems stuck in the queue/process stage.

The past couple days I have also seen this happen where a backtest will get stuck for a long time. I believed it was due to my oversized universe & heavy use of History or some init bug I had introduced. I've see this behavior on the public and dedicated QC cloud. Perhaps related to the History service?

Sometimes when I run a backtest, it will report the usual ram usage; for example let's say 2-4 over its backtest when charted. Then the exact same algo run again(just hit backtest again) will seem to mem leak and use 4-40+ gigs of ram after a slow startup. Could this odd behavior be related? I have seen this mem issue before the "stuck" issue appeared the past couple days.

Thanks Derek, Charles sorry about the issues. Demand has surged 300% in the last 3mo so we've had to reengineer our cloud architecture. We've come up with a nice scalable solution and it'll be live this week we hope. We're just waiting for the data to copy to the new architecture.

The new architecture will have reliable speeds (be predictable) but will be about 20% slower than current design when no one else is online. The reality is with 100-200 people online the current architecture starts to crawl so we hope the average speed will increase with the new design.

Also; due to limitations of the new cloud (AWS) we'll have RAM limits of 16/32gb max. We'll keep pondering the way around this especially as we want to support your case Derek. 

Thanks Jared for the info!!

I have been spending a lot of time learning Quantconnect mostly because it connects with my broker and I hope to stay here!  There have been lots of challenges so far for me and I might be able to look past that but the speed part added to waiting forever to find out there are erros sometimes, and then trying to figure out what the error could possibly be... is over the top. Looking forward to any changes that will help with the speed for sure.

100-200 people online sounds very small actually considering how rapidly the quant universe is growing, What would happen when you get into the thousands of users?

Does any of this affect live trading algorithms?

FYI my current algo I just tested ended up taking 625.44 seconds to deploy and hasn't even started to show trades yet. There are 129 users online.

When you say how much work you do on start, does that mean how much code goes in during the Initialize code?

If so - is it better to do more "work" in that code, or less for deploy speed?

Yes, that is what it means. For instance you could be adding a lot of securities and making History Data Requests to update indicators. In that case, it would be better to rely on SetWarmup.