Re: [amibroker] Re: Quad-core test results, AmiBroker Email List Archive

Paul,

I?ll need some time in a dark room before I?m ready to start thinking about getting IO to utilize multiple cpu?s / cores on a single machine but the principles should be the same as getting a cooperative task across several userid?s to work using scripts/programs and AB?s standard Exhaustive Search mechanism.

So ? Before I even think about going where IO lives I?ve taken the short proof of concept route.

Given the benchmarks that have already been run, silly or otherwise, it would seem that this is worthwhile in a lot of cases and will only become more so as time rolls along and I already have this working in rough form with no synchronization ( the grunt work ) after a couple of hours of ?play? ?

The PITA in doing this outside of AB is of course that things need to be synchronized between the different running copies of AB i.e. Settings / Database / Reference Ticker ? on and on and on ? i.e. the kind of things that IO needs to ensure are the same between the client and the server(s) but doing this on a single box is still simpler because for example there?s no database or portion thereof that needs to be cloned from the client to the server(s) etc but it?s still a PITA.

You get a big thank you for the idea as I had no clue this was even possible using multiple userid?s but I wish you had mentioned it when you first discovered it as I suspect I and a lot of other people could have saved a lot of CPU cycles over the last couple of years ? LOL ?

Fred

From: amibroker@xxxxxxxxxxxxxxx [mailto:amibroker@xxxxxxxxxxxxxxx] On Behalf Of Paul Ho
Sent: Saturday, May 24, 2008 11:56 PM
To: amibroker@xxxxxxxxxxxxxxx
Subject: RE: [amibroker] Re: Quad-core test results

Fred

A couple of questions

1. Can runas be used on IO ie IO for use1 is connected to AB for user1 and IO for user2 is connected to AB for user2?

2. regarding the problem of multi-core CPU each running a copy of AB bearing down on a single IO problem. You are today doing it with IOserver running on separate machine. Cant you treat multi-core CPU as separate virtual machine. ie IOserver would be runned on one core (actually as different user) accepting command from IO from another core(user) through the TCP/IP layer?

From: amibroker@xxxxxxxxxps.com [mailto:amibroker@yahoogroups.com] On Behalf Of Fred Tonetti
Sent: Sunday, 25 May 2008 2:40 AM
To: amibroker@xxxxxxxxxps.com
Subject: RE: [amibroker] Re: Quad-core test results

Paul,

While this method works, I would think it would be simpler to utilize something like the RunAs command as this would allow all sessions to be viewable from a single console w/o RDC or VNC etc i.e.

Assuming there are userid?s IO1 through IOn defined then

RunAs /profile /user:IOn ?c:\...\Broker.exe?

See the attachment ?

Beyond this it doesn?t directly deal with the issue I really wanted to which is to bring multiple cpu?s / cores from a single box to bear on a single problem using the features of an intelligent Optimization. It does with AB alone using an exhaustive search approach since this type of optimization problem can be easily split into logical pieces and conceivably scripts could be used to initiate the processes under each of the userid?s and gather the results from them into a single place. I had assumed for the last couple or three years that multi cpu / core utilization for optimization in AmiBroker would be forth coming relatively soon but since that hasn?t materialized maybe the best approach then is to use the tools that already exist. Of course I?m sure if I spend the effort to do this that like magic the functionality will suddenly appear where it more properly belongs. Then again I suppose I should have just done this several years ago.

From: amibroker@xxxxxxxxxps.com [mailto:amibroker@xxxxxxxxxps.com] On Behalf Of Paul Ho
Sent: Saturday, May 24, 2008 1:38 AM
To: amibroker@xxxxxxxxxps.com
Subject: RE: [amibroker] Re: Quad-core test results

Fred and others

How to setup multiple instances of IO/AB on Quad Core:

Since this might be of interest to a few. I would post it here.

This is relatively straight forward, the instructions below applies to XP and XP64. I dont use Vista so I dont exactly know, but I expect it to be similar

1. Create multiple users accounts in XP, each user account will be used for each instance of IO. You might decide to have a separate database for each user, but it is not necessary. What is necessary is to setup IO on a different path for each user to avoid overwriting the same files. I do that with the /IOpath switch as an argument. For others who are interested in doing that refer to Fred's doc. You can optionally open up task manager and set process affinity to a particular core. I found that unnecessary.

2. loggin in. - There are 2 methods of logging in. Via console or remote desktop connection. For the first user, just login in as usual, bring up AB and start IO. Once AB/IO is running. You have move onto the 2nd user. Since XP by default only allows 1 active user at a time. We have to "log out". If you are at the console. It is important that you log out using switch user. otherwise all your jobs will be killed. If you are using remote desktop, you actually dont log out but disconnect by clicking the "x" button at the top. Alternatively, if you're using XP32, you can download a patch on the internet that allows multiple log ins, which means no disconnection is necessary.

3. 2nd and subsequent log in. Similar to the first.

4. Monitoring multiple instances. if you have downloaded the patch and are able to run multiple remote desktop connections. YOu can easily monitor the progress via the connections that you have. Unfortunately,There is no patch for XP64, So I am left with switching between users every now and then. But I usually mount the IO directory as a network drive from another computer. so I'm able to at least monitor two copies of IOstatus.

From: amibroker@xxxxxxxxxps.com [mailto:amibroker@yahoogroups.com] On Behalf Of Fred Tonetti
Sent: Friday, 23 May 2008 11:36 PM
To: amibroker@xxxxxxxxxps.com
Subject: RE: [amibroker] Re: Quad-core test results

Paul,

Interesting and no I can?t say that I have looked into it.

I would think the drawback to this approach would be that while this very well might work as you say, in an intelligent optimization the work being done by each IO / AB / (CPU / Core) combination cannot really be cooperative per se. This is because unlike an exhaustive search there is no way to logically split an intelligent optimization except on a generation by generation basis which implies there would have to be communication and/or control between each generation i.e. in order for it to be cooperative I would think there would need to be a higher level piece of software controlling what was happening with each IO / AB combination ? or ? there would need to be an IOServer like program running for each CPU / Core beyond the first one that gets treated like a sort of local server to the client.

From: amibroker@xxxxxxxxxps.com [mailto:amibroker@xxxxxxxxxps.com] On Behalf Of Paul Ho
Sent: Friday, May 23, 2008 2:37 AM
To: amibroker@xxxxxxxxxps.com
Subject: [amibroker] Re: Quad-core test results

--- In amibroker@xxxxxxxxxps.com, Fred Tonetti <ftonetti@xx.> wrote:
>
> PS .
>
>
>
> I have noticed basically the same thing on dual core and core 2 duo
type
> machines i.e. the same optimization on a single symbol can be run
> simultaneously on both cores with no increase in run time .
>
>
>
> This seems to apply as well to optimizations on hundreds of
symbols ..
>
>
I think it all boils down to L2 Cache size, certainly the more
symbols + the longer the data series, the larger L2 cache size has to
be. Certainly I notice a marked increase in run time with both cores
running when the machine is loaded up with symbols and/or with long
data series vs few symbols and truncated data series.

>
> While the tests that TJ performed seemed to imply otherwise, the
implication
> of the above would seem to be that it would be beneficial for AB to
do the
> heavy lifting in terms of getting multiple cpu's/cores involved in
an
> optimization when they exist so that results could be combined in
one place.
>
>
I have been able to run multiple instances of IO, one on each cores
for about 2 years. All I need to do is to create a separate user
login per instance. I always thought that it was possible for IO to
treat multiple users on a single machine as multiple machines and use
TCP/IP services to communicate. Have you looked into this possibility?

>
> If this were the case then IO would also take advantage of this on
> individual machines as well as on multiple machines simultaneously.
>
>
>
> _____
>
> From: amibroker@xxxxxxxxxps.com [mailto:amibroker@xxxxxxxxxps.com]
On Behalf
> Of Steve Dugas
> Sent: Thursday, May 22, 2008 8:50 PM
> To: Yahoo - AmiBroker
> Subject: [amibroker] Quad-core test results
>
>
>
> Hi All,
>
>
>
> I finished setting up the new quad-core machine and ran my first
tests
> today, so am posting the promised test results. All I can say is
Wow!
> Really nice improvements, even better than I expected to see in my
highest
> hopes. I am happy as a pig in sh*t !! 8 - )
>
>
>
> For background, the test was an optimization run on a single
ticker, which
> is how I will be running all my other tests ( ticker was QID, 468
EOD data
> bars ). All tests used the same code with different param settings,
all
> tests had about 46,000 opt steps.
>
>
>
> To put things in perspective and show why I am so happy, the
original code
> was about 2,400 lines. The first time I ran it, on my old backup
computer
> which I was using as a dedicated optimization machine, it took 7
1/2 hours
> to run. So then I copied the code and created a shorter version,
removing
> everything which wasn't absolutley necessary for the optimizer, and
that
> reduced the run time to 2 1/2 hours. Then I ran this short version
on my
> faster primary machine and that took 1 1/4 hours, which is about
what I was
> expecting to see on the new machine.
>
>
>
> So today I started by running just ran one instance on the quad-
core - that
> took only 30 mins and Task Manager showed it was using just 25% of
the total
> processing power! Well to make a long story short, I kept adding
one
> instance at a time, all instances ran in 30 mins and each used up an
> additional 25% of the CPU power. In the end, I was easily running 4
> simultaneous instances. This pretty much kept the CPU tached at
100% but all
> instances ran fine, all finished in 30 mins and I didn't experience
any
> problems at all. I was even saving the first ones to spreadsheets
while the
> final ones were still finishing, wow everything just worked
flawlessly! So,
> the quad-core can run 4 seperate opts simultaneously in 30 mins,
which
> averages out to 7 1/2 minutes per opt, which =
>
>
>
> 10X improvement over running the short code on the fast machine...
>
> 20X improvement over running the short code on the old dedicated
> optimization machine...
>
> 60X improvement over running the original code on the old
optimization
> machine... Awesome !!!
>
>
>
> To those who were wondering what is the best machine to get for
running AB,
> it looks to me like quad is the way to go. ( My machine has an
Intel
> processor, which TJ mentioned should probably work better than AMD
for this
> stuff )....
>
>
>
> Steve
>
>
>
>
> _____
>
> I am using the free version of SPAMfighter for private users.
> It has removed 459 spam emails to date.
> Paying users do not have this message in their emails.
> Try SPAMfighter <http://www.spamfighter.com/len> for free now!
>

I am using the free version of SPAMfighter for private users.
It has removed 459 spam emails to date.
Paying users do not have this message in their emails.
Try SPAMfighter for free now!

I am using the free version of SPAMfighter for private users.
It has removed 460 spam emails to date.
Paying users do not have this message in their emails.
Try SPAMfighter for free now!

I am using the free version of SPAMfighter for private users.
It has removed 460 spam emails to date.
Paying users do not have this message in their emails.
Try SPAMfighter for free now!

I am using the free version of SPAMfighter for private users.
It has removed 460 spam emails to date.
Paying users do not have this message in their emails.
Try SPAMfighter for free now!

I am using the free version of SPAMfighter for private users.
It has removed 460 spam emails to date.
Paying users do not have this message in their emails.
Try SPAMfighter for free now!