|
Concurrent Historical Tests |
LiamsDad
|
Post subject: Concurrent Historical Tests |
Post rating: 0
|
Posted: Thu 24 Mar, 2016, 13:29
|
|
User rating: 0
Joined: Wed 10 Jul, 2013, 12:42 Posts: 30 Location: FranceFrance
|
I have a question regarding Visual JForex and Historical Testing. I have been teaching myself Visual JForex and I have a couple of ideas I want to backtest and tweak against various currency pairs.
To save time I would like to backtest in parallel, ie run historical testing against GBPUSD, EURUSD, AUDUSD and USDJPY all at the same time but with a lot of data, ie against 1 year tick data. The idea is to run them on virtual machines in the cloud as all this data will take time to process. Currently I have 3 virtual machines set up in the cloud running 1 historical test per server but after running OK for an hour or so, server 2 and 3 slow down to a crawl ie slower than real time 1x speed.
All of these are separate servers running the same VisualJF Interface account against the Visual JF demo account.
Is there any backwards and forwards processing in the historical testing process, or is test data ie tick data Jan-1-2016 to Feb 29 2016 downloaded and run locally? I am trying to determine where the bottleneck is - client/server processing or perhaps the server cpu is at fault.
It is required/recommended that I get separate demo accounts to run these historical tests in parallel?
Regards LD
|
|
|
|
|
amine_chourou
|
Post subject: Re: Concurrent Historical Tests |
Post rating: 0
|
Posted: Wed 30 Mar, 2016, 15:13
|
|
Visual JForex expert at Dukascopy | |
User rating: 288
Joined: Thu 18 Apr, 2013, 09:26 Posts: 1496 Location: Switzerland, Geneva
|
Hi,
Visual Jforex historical tester call the same data as Jforex platform for testing purpose but the engine behind it is less powerful compared to Jforex. Besides, it is much better to test on 6 months basis (or less) especially when it comes to tick data (If your strategy needs concretely tick prices and there is no way to optimize it). I don't think that using separate demo accounts will speed up the historical tester substantially because the data load is mainly at the level of your Java process which remains the same regardless of the number of accounts used. I recommend the use of 6 months tick data test in Jforex platform for instance and check the result.
Thanks
|
|
|
|
|
LiamsDad
|
Post subject: Re: Concurrent Historical Tests |
Post rating: 0
|
Posted: Wed 30 Mar, 2016, 15:49
|
|
User rating: 0
Joined: Wed 10 Jul, 2013, 12:42 Posts: 30 Location: FranceFrance
|
Thanks for the reply Amine
For other forum users thinking of using cloud based Virtual machines for historical testing:
I have tried processing using the Amazon AWS free tier micro T2 level running Windows Server 2012 and Visual JForex historical tester stalls to a crawl in about an hour, even with only 30 days of tick data.
I have since switched to Google Cloud platform which offer a 60 day free trial, provisioning a 2 Core Xeon VM and a 4 Core Xeon VM both using Windows Server 2012 for processing Historical tick data. 2 Core server maxes out CPU processors at 100% and slows down considerable (which was a surprise to me), 4 core server performs ok. Going forward I will see if I can optimize that 4 core baseline.
Google Cloud offers 60 days free trial of their Compute Engine, offering 8 cores total so I'll split into 2 x 4 core servers running separate demo accounts for historical tests and evaluate the cost/benefit compared to buying/provisioning my own server after the 60 days.
If anyone else has any experience with using cloud VMs for historical testing I'll like to hear others experiences.
LD
|
|
|
|
|
|
Pages: [
1
]
|
|
|
|
|