TYCanadian
I don't have nearly the requirement that you
do or the platforms to test these on,
but I have a 3.4 GHz hyperthreading
Intel board with paired 2 gig of memory and SATA high speed disks;
and what I see is that in the two HT areas
shown on the XP Windows Task Manager, my processor utilization
only runs up to 50%, rarely 60% on each
scaling up and down between 30-50% on either.
From that my thinking, the bottle
neck is the data access extends the run times and throughput.
One example is running an optimization
on of a trading routine on a single stock index the
NDX100 and
then running the same job on 100 stocks in
the NDX100. The time for convergence on the
IO optimization
on the NDX100 runs up over 1/2 hour with the
multiple stocks.
So I don't think the bottle neck is the
processor, memory or caching? but I'm no expert. I think data base access
through plugins.
and the plug in caching is the pacing
processes- even though I've got one of the fastest plugins according to TJ for
the QP3 database.
Best regards
JOE
----- Original Message -----
Sent: Wednesday, June 22, 2005 4:10
AM
Subject: [amibroker] AmiBroker Computer
Performance -- What Matters?
Hi,
I am doing a few studies of my trading system
that require a LOT of computing power. I need to run a few thousand
optimization steps. I know CPU matters, but I'm wondering what
specifically matters (FLOPS, MIPS, Cache size, etc.), as well as what
other factors make a big difference (RAM, Hard Drive speed, etc.).
Basically, I'm curious what to emphasize or expect if I were to obtain
some new PC's.
Normally I would just test and see what makes a
difference, but I am a bit baffled by what I am seeing. My stock
database is roughly 560 MB. Here are some system specs along with
backtest times for my system:
P4 1.8 GHz, 1 GB RAM, 10,000RPM SCSI
Ultra 160 -- 8.5 minutes P4 1.8 GHz, 512 MB RAM, 10,000RPM SCSI Ultra 160
-- 9.2 minutes P3 0.733 GHz, 768 MB RAM, 10,000RPM SCSI Ultra 160 -- 9.6
minutes P3 0.733 GHz, 256 MB RAM, 10,000RPM SCSI Ultra 160 -- 9.8
minutes Dual P4 Xeon 3.06 GHz, 2.5 GB RAM, 10,000RPM SCSI Ultra 320
(running two simultaneous instances of AmiBroker, two backtests
simultaneously) - - 4.1 minutes (total time for a total of two
backtests)
Basically, I am wondering why the P4 with 512MB RAM barely
outperforms the P3 with 256 MB RAM, but the Xeon system is much faster
than the P4 systems. Is hard drive bandwidth really that big of a
factor? Both the P4 system with 1GB RAM as well as the Xeon system
seem to grab all the RAM they need (with some still left over).
Any
thoughts on this would be welcome.
Thanks.
Please note that this
group is for discussion between users only.
To get support from
AmiBroker please send an e-mail directly to SUPPORT {at}
amibroker.com
For other support material please check also: http://www.amibroker.com/support.html
Please note that this group is for discussion between users only.
To get support from AmiBroker please send an e-mail directly to
SUPPORT {at} amibroker.com
For other support material please check also:
http://www.amibroker.com/support.html
Yahoo! Groups Links
|