A few comments on how to maximise cloud power
GSB benchmark allowed me to tweak my cloud settings to maximise speed.
https://trademaid.info/gsbhelp/GSBBenchmark.html
I have been observing cloud behavior a bit.
Some users workers are chewing up 28 gb each.
This is due either to using 29 30 31 and data2 58 60 62 minute bars
(my own tests on ES showed this was not helpful)
or verification on anther 8 markets is done at build time.
I think verification at build time is not a good idea as its much more cpu intensive, and should be done later in the build process. This is how the
methodology is done.
https://trademaid.info/gsbhelp/Methodology.html
Im tempted to remove the verification worker feature, or remove it from free cloud.
So here is the implications of using lots of ram.
1) Resource managers are set to terminate all workers that use over x gb of ram.
We need this as some people who donate to the cloud have say 16 gb of ram, if you try to get 28 gb...crash. So a portion of your workers on the free
cloud will get terminated.
2) Resource manager terminated the worker that uses the most ram if the computer runs out of ram. So again you will get terminated and another user
take your place on the cloud.
GSB is great for testing our assumptions.
One assumption here is that what doest work so well on ES, doesnt work well on other markets.
So in this case you need to build the 50,000 systems on 28,30,32 with data2 59,60,61 and see if its producing better stats in favourites c,d,e,f just
like appengineer did posted on 14-9-2019 at 03:01 PM
here https://trademaid.info/forum/viewthread.php?tid=117&page=11
(he correctly followed the methodology path)
One additional factor why ram is high, is Ive set the cache settings to be large which gives most speed. Im peaking 9 GB ram usage per worker using
es 29,30,31 bars
and 4 GB during WF doing 3 WF at once.
|
![[*]](images/davis/default_icon.gif){kind=icon}
