This location cannot be changed. Additionally, some of the results are also stored within the ST_DATA.MDB file.

Thanks for replying that fast. Unfortunately, my idea won't work. My point was to use a "ram drive" in order to increase the speed of the EST. I thaught that limiting the number of acces to the hard drive would increase the processing speed. By the way, is there a way to "pause" the system tester in the middle of a testing, and to resume it later? Thanks again for your help. John

h, you're right.... John, Put the data onto the fastest disk you have (RAM drive = = better!) Accessing the data is the slowest part of the system test, second only to poor coding! If you store the data in the fastest media possible, the test will be faster. Writing the results to the .mdb is inconsequential, but if you MUST have this fast too, then you are going to have MS running from the ram drive too. Hope this helps. wabbit :D P.S. Wanna give us a review of your RAM drive? Who, what, when, why, how, how much, caps and lims etc?

Hi there! Here are the first results of my testings. And..... WOW! It seems that by simply working on the MS Data that is located in the ram drive increases the speed of the EST by +/-75%!!! These are only tentative results. But as soon as I will be sure of the improvement, I will post again here (let me 2 to 3 days :) ). And guess what... If all this is confirmed, the next step will be to install MS in the ram drive. This should not be a big deal in fact. I just have to back up the ram drive content onto the hdd before switching of the computer. More to come next week!

Wabbit, In fact I am just working on one security at a time. I am testing various indicators on indexes (i.e. SPX, NKY, FTSE,...) using long time series (5 to 20 years). So I do not have the problem of using an unuseful large universe of stocks. Why am I doing this? To demonstrate to my "students" that you cannot use a specific indicator with specific parameters and expect that it will give you accurate signals during 10 years (in other words, things change and that you have to redesign you trading system on a regular basis). Concerning the ram drive. In fact I am not using an external ram drive (i.e. some kind of hdd made of DRAM). What I am using is an emulation of a drive within de main memory of my computer. I only created a small drive of 64MB in the main memory of my computer to store the MS Data folder. To do this, I use a small freeware called AR RAM Disk. It only works with Win NT/2k, but I am sure that it must be possible to do the same on other OS. That's why I am experiencing such a dramatic improvement. No more PATA or SATA interface. I am just enjoying the full bandwith and access time of my modest DDR. If I can confirm the results, this might be a good indication for the developpers to improve the speed of MS. If the amount of data is not to big compared to the available main memory, it would be a good thing if MS was loading all the data in the ram and working on it from there. I hope that all this is clear (english is not my mother tongue) and that I answered to your question. I go back to my tests and will give you some more news next week. John

Hi Wabbit! Ok, here are the final results of my tests, the harware I am using and the procedure I used. First, the hardware and software (it's not the latest and fastest on the market). AMD Athlon 2600+ 1,5GB DDR HDD 80GB, 2MB cache, 7200rpm, PATA. O/S: Win2K MS 9.0 The procedure. I tested different indicators (RSI, Stochastics, MACD,...) one at a time, on one index (SPX, NKY,...) at a time. I used daily times series over 20 years and used from 3 to 4 optimizing variables. The number of iterations for each system was between 100'000 and 1'000'000 and I stored the 50 best full results (no summary results). Initially, all the data was stored on the hard drive. Then, I copied the source data (i.e. the MetaStock Data folder) to my virtual RAM drive. This decreased the time needed to complete the system testing by +/-40%. I then decided to install MS in the virtual RAM drive in order to have no more access to the HDD. And the result is even better: the time needed to complete a system testing is decreased by almost 90% (i.e. 10 times faster). Part of this dramatic improvement is surely due to the fact that I have a slow HDD. If I had a RAID 0 array based on SATA or SCSI hard drives, no doubt the difference would have been less impressive. However, I think that if you are using a lot of source data (long time series, lots of stocks,...) and if you are storing extensive results, the virtual RAM drive will improve the speed whatever the HDD you are using. Furthermore, based on what I understood on your own tests, I think that we did not use the same procedure and amount of data to retreive and to store. That's probably another reason why you results and mine are so different. I'd like to have your opinion about that. Thanks a lot. John

Wabbit, I have Norton AV 2005 running with max protection activated as I am connected to Internet all the time. However, it should also scan all the files coming from and going to the virtual RAM drive (I guess). None of my folders or disks are shared. I have the historical data of the DJI from 01/01/1929 until today. I think that this sould be enough to compare our results. What I suggest you is to run a test from 01/01/1929 until 31/12/2006 on the DJI. The system to test would be the following (I will mention the tabs from the system editor): -General Long Orders Single Bias 1 simultaneous position -Buy Order cross(rsi(opt1),op2) -Sell Order cross(opt3,rsi(opt1)) -Sell short, Buy to cover, Stops nothing -Optimizatons opt1 from 1 to 50 step 1 opt2 from 5 to 50 step 5 opt3 from 50 to 95 step 5 (this should give us 5000 iterations which will be enough to be able to compare our results without running for hours) The test I will be running will be a "point only test" with long only positions. I will store the 50 best results. This wil NOT be a quick test. All the other parameters are left blank or to their default value. I will run this test 6 times: -first time using MS installed on the HDD, all data coming from and going to the HDD. -second time using MS installed on the HDD, all data coming from the virtual RAM drive and going to the HDD. -third time using MS installed on the virtual RAM drive, all data coming from and going to the virtual RAM drive. -fouth, fifth and sixth times: same as above but with Norton AV 2005 desactivated. This way, we will see how the virtual RAM drive improves the running times under different situations. Let me now if this is ok for you. If there is some other test that you would like to run, I will do it. John

Wabbit, Let's change slightly the rules. I will only store the best 20 full results, as 50 full results is a bit too large for my RAM drive. The period for the test will be from 01/01/1929 until 30/12/2005 (and not 31/12/2006 :oops: ). John PS: BTW, what is the generic RAM drive that you used? What is your OS?

John, I have done some of my testing with some surprising results. My system: P4 2.4GHz 1GB RAM 120GB 7200rpm 8MB Cache HDD Windoze XP Pro SP2 PCCillian AV and Firewall Test1: Long entry and exit criteria as discussed. 01/01/1929 until 30/12/2005. Points only test. NOT a quick test. Store the best 20 results. MS installed (as normal) on HDD DJI data on HDD Rationale: this one of the original test prospects. Test2: Long entry and exit criteria as discussed. 01/01/1929 until 30/12/2005. Points only test. NOT a quick test. Store the best 5000 results. MS installed (as normal) on HDD DJI data on HDD Rationale: this is to test whether it takes longer to write to the ST_DATA.mdb file if there are more results to store. It shouldn't take any longer to write to a Jet Database containing 1 record or 10000 records, that's the whole idea of databases! Test3: Long entry and exit criteria as discussed. 01/01/1929 until 30/12/2005. Points only test. NOT a quick test. Store the best 20 results. MS installed in RAM Drive DJI data in RAM Drive Rationale: one of the originaly propsed tests. Test4: Long entry and exit criteria as discussed. 01/01/1929 until 30/12/2005. Points only test. NOT a quick test. Store the best 20 results. MS installed (as normal) on HDD DJI data oin RAM Drive Rationale: one of the originaly propsed tests. Conditons: With the exceptionof Test2, each test was run with a completely empty ST_DATA.mdb (with the exception, of course, of the system test being performed). It contained no results of any kind. Test2 was conducted with the ST_DATA.mbd containing the 20 results from Test1. The RAM Drive was created and maintained using a trial version of RAMDiskXP from www.cenatek.com The disk was 100MB and was carefully monitored during the testing to ensure it was never completely 'full'. There were a couple of issues trying to get MS installed in this space, as standard installation (and installations when not paying full attention) also includes the .wmv files, .pdf instructions, .doc advertising etc which are all extraneous to the requirements of running MS. Once I had cut down the fluff, we were left with the smallest working instalation of MS. Results: During the running of each test, I noted the progress of the test and plotted these in Excel to predict the finishing time for each test and to provide a graphical representation of all of the tests versus each other test. During the progress of each test, I wa able to see how closely each test 'fitted' to the predicted performance. If the correlation was close to my estimates, I stopped the test at 50% complete to save my time and sanity. The assumption made is that the performance of each of the tests is the same in the second 2500 tests as the first 2500 tests. I could not see any reason why this would no be the case, but there is one issue that I will mention shortly. The results surprised me. There was no apparent speed advantage using a RAM Drive over standard physical memory, IN MY COMPUTER! During the tests involving the RAM Drive I observed the HDD access was significantly less than when the data/application is being taken from the HDD. The accesses 'seemed' appropriate with each of the tests being conducted. Attached is the Excel graph and the raw data: Notice that all the tests conducted would have completed in just under three hours. This is regardless of wheher the application is running from the HDD or RAM, and the data. Why is this so? Professor Julius Sumner Miller made this quote famous in Australia, but it is still very relevant today as we ask the same question of just about everything we do. I believe the speed of the EST is dependent on a lot more than just MS. I believe the Operating System has a lot to answer for in this regard. The cache control imposed/implemented in my version of Windows XP needs further investigation? Correct. Lets note, the .mdb is a Jet Database, yet the results files are .dta files located in folders 1-10 under folders 1-10, so there is 100 folders containing the full system results. Databases are supposed to have dymnamic data coming to and going from the database on a regular basis, however, writing the results to a .dta file seems to be the biggest slowdown in the entire process. This is the reason that "Quick Tests" are sooo much faster than full system tests. If Equis were going to redevelop the EST, this is the part where they need to spend some time? The earlier improvemtn did seem a bit strange, but it was not beyond the realm of possibility. I agree the source data is 'uploaded' into RAM for rapid access, but the full affect of this process and the efficient RAM employment is not going to be visible when running a test on a single security. To really test out the spec, I suggest that and entire market be subjected to the test. My data file for all ASX equities is about 69MB, so installing this small amount of data when I have 1GB of RAM (less OS overheads and RAM Disk space) is going to be insignificant. To stress the system, we need a lot more data and a lot less RAM! Personally I think it would be more efficient to run the entire test, storing the results in a temporary area or RAM or physical memory in the order in which they occurred. At the end of the test, sort the results and return the required values. This will cut down on processes, but increases the memory requirement for the tests. This would of course mean the results writing procedure would have to be seriously modified also! My CPU was at or very nearly at 100% for the duration of the test. The was the occasional dip to about 98% but this did not occur regularly, nor did it seem to correspond with any particular progree of the EST tests. AAARRRGGHH - Damn phpBB trimmed my post!

Where's your Excel spreadsheet attachment?

And what the heck happened to the "Follow thread" and "Favorite thread" options?