John Colby
jcolby at colbyconsulting.com
Fri Oct 17 11:13:00 CDT 2003
>I've made two posts to this effect in the past week and have seen neither of them show up yet. Just goes to show, "processor speed is rarely the bottleneck". <grin> John W. Colby www.colbyconsulting.com -----Original Message----- From: accessd-bounces at databaseadvisors.com [mailto:accessd-bounces at databaseadvisors.com]On Behalf Of Henry Simpson Sent: Friday, October 17, 2003 10:33 AM To: accessd at databaseadvisors.com Subject: RE: [AccessD] Number vs text data type A person I know upgraded his computer on a 2 person LAN with a machine three times as fast. His data was kept on his secretary's computer on a 100 mbit LAN connnection. I warned him that he wouldn't likely see significant difference in performance. In fact, he could see no difference. I had warned him that a bit of time looking at indexes and addressing data volume and recordset size would probably be of more benefit and when all was said and done, he came asking for help. Processor speed has rarely been the bottleneck in most Access deployments. I've made two posts to this effect in the past week and have seen neither of them show up yet. The last was over 36 hours ago and I've seen once response to it that included an excerpt of my post. I wonder what fraction of all posts made to the list that I ever see. I tried looking at Drew's archive but if I search on my name, the last post by me is August 30, 2003. I'll send this once per day for the next week until I see it show up. Hen >From: Drew Wutka <DWUTKA at marlow.com> >Reply-To: Access Developers discussion and problem >solving<accessd at databaseadvisors.com> >To: "'Access Developers discussion and problem solving'" ><accessd at databaseadvisors.com> >Subject: RE: [AccessD] Number vs text data type >Date: Wed, 15 Oct 2003 14:11:41 -0500 > >You are right, Zip Codes could change into anything. But that applies to >everything we know. Should we make EVERY field a string? You can only >code >for the present, and predicted future. The post office isn't going to >change the Zip Code system (at least not the basic 5 digits) without a >decent amount of lead/warning time. They have systems they'd have to >modify >also. > >Also, you are right, with today's processors, speed is not as critical. >But >the question is, when faced with a factor of 1.5x to 2x faster, do you go >with the slower process, or the faster process? Isn't it better to stick >with the faster process? Why sacrifice any speed, to accomodate unknown >future variables? > >Drew > >-----Original Message----- >From: Heenan, Lambert [mailto:Lambert.Heenan at aig.com] >Sent: Wednesday, October 15, 2003 1:47 PM >To: 'Access Developers discussion and problem solving' >Subject: RE: [AccessD] Number vs text data type > > >Ah but... > >one day in the future the design of zip codes (which is only ONE example of >a "number" that is not a number) could well be changed to something utterly >different that looks nothing like a number. You can't just keep adding >fields and complex put-it-all-back-together routines to reconstruct the >original string from a combination of strings and numbers. Well you can, >but >it might make life miserable as "the management" make more wacky decisions. > >As for searching for string vs numbers. Here are some results on a test Db >I >threw together. > >Using a table with 581,672 records: each record has a longint field and a >string field. The long field is populated with random long integer values, >and the string field has the string representation of its corresponding >integer field. Both fields are indexed with duplicates allowed. > >The table is in an MDB file (44.9 MBytes) which is on a network drive, and >the network is running at just 10-mbit. So then I ran the following code to >search for random 'numbers' both as numbers and as strings... > >Sub testSearch() >Dim n As Long >Dim s As String > >Dim TextSearchStart As Single, TextSearchEnd As Single >Dim NumericSearchStart As Single, NumericSearchEnd As Single >Dim TextSearchTime As Single >Dim NumericSearchTime As Single >Dim x As Long >Dim strHits As Long >Dim intHits As Long >Dim db As Database >Dim rs As Recordset >Dim sCriteria As String > Randomize > Set rs = OpenDynaset(db, "tblHuge") > For x = 1 To 1000 > With rs > n = CLng(Rnd() * 2 ^ 31) > s = CStr(n) > > sCriteria = "nNumber=" & n > NumericSearchStart = Timer > .FindFirst sCriteria > intHits = intHits + Abs(CLng(Not .NoMatch)) > NumericSearchEnd = Timer > NumericSearchTime = NumericSearchTime + (NumericSearchEnd - >NumericSearchStart) > sCriteria = "sText='" & s & "'" > TextSearchStart = Timer > .FindFirst sCriteria > strHits = strHits + Abs(CLng(Not .NoMatch)) > TextSearchEnd = Timer > TextSearchTime = TextSearchTime + (TextSearchEnd - >TextSearchStart) > End With > Next x > rs.Close > Debug.Print "Average find times" > Debug.Print "Numbers:" & Format(NumericSearchTime / 1000, >"0.0000000000") & " Hits:" & intHits > Debug.Print "Strings:" & Format(TextSearchTime / 1000, "0.0000000000") >& >" Hits:" & strHits > Set db = Nothing > Set rs = Nothing >End Sub > > >And here are the results of a few runs, last to first... > >Average find times >Numbers:0.0020859370 Hits:0 >Strings:0.0025468750 Hits:0 >Average find times >Numbers:0.0023203120 Hits:0 >Strings:0.0030234380 Hits:0 >Average find times >Numbers:0.0023984380 Hits:0 >Strings:0.0040195310 Hits:0 >Average find times >Numbers:0.0029257810 Hits:0 >Strings:0.0046445310 Hits:0 > >As you can see the random search values never generated any hits - not too >surprising considering the number of possible values in 32bit long. > >The numeric search times were fairly consistent, and the string search >times >steadily dropped. I assume that there is some sort of caching going on >here. >But the main point is that at worst it took just about twice as long to >search for the string as it did to find the long. Not *such* as huge >performance hit considering we are talking about thousandths of a second. >There is also a fair bit or arithmetic going on between each search. > >I goes without say (so why am I saying it?) that whne yo take out the >indexes on the fields the search times go though the roof (for the Long >field as well as the string one). > >Lambert > > > -----Original Message----- > > From: Drew Wutka [SMTP:DWUTKA at marlow.com] > > Sent: Wednesday, October 15, 2003 12:26 PM > > To: 'Access Developers discussion and problem solving' > > Subject: RE: [AccessD] Number vs text data type > > > > Ah, but you can always add fields to handle new anomalies! With the zip > > code, just add an 'extension' field. Original Number field is left >intact > > (and all 'search' routines) would still work. > > > > Drew > > > > -----Original Message----- > > From: Heenan, Lambert [mailto:Lambert.Heenan at aig.com] > > Sent: Wednesday, October 15, 2003 8:38 AM > > To: 'Access Developers discussion and problem solving' > > Subject: RE: [AccessD] Number vs text data type > > > > > > This thread has been going on quite some time now, so why should I let >it > > stop :-) > > > > One key issue that which has not yet been mentioned but should be > > considered > > is is there any possibility that the design / structure of the >"numerical" > > identifier might change? > > > > To take the example of US Zip codes. Once upon a time there were no Zip > > codes, then along came 5 digit zips, and a little later the hyphen plus > > four > > digit extensions. This kind of change could happen at any point in time > > because at the end of the day a postal code is an *arbitrary* >identifier. > > In > > other words there is no arithmetic meaning in a zip code, so it should >not > > be stored as a number. You CAN store a 5 digit zip code as a Long Int, >but > > you'll regret it if "the powers that be" decide to change things later >on. > > > > As for searching for strings in a sorted list, while it is slower to > > compare > > a whole string with another whole string, vs. comparing numeric values, >to > > search a sorted list really does not take that long as a standard binary > > chop approach is going to zero in very quickly on the target string. (If > > my > > mental arithmetic is up to it, searching for a string in a list of one > > billion (10^9) strings would require 30 comparisons at most.) In >addition, > > the initial stages of the binary chop only need to compare the first > > character (or first few) in the string with the first character (or >first > > few) of the target, not much slower than integer data comparisons. And > > it's > > just as simple to figure out if a target is within some range. > > > > > > > > Lambert > > > > > -----Original Message----- > > > From: Drew Wutka [SMTP:DWUTKA at marlow.com] > > > Sent: Tuesday, October 14, 2003 5:18 PM > > > To: 'Access Developers discussion and problem solving' > > > Subject: RE: [AccessD] Number vs text data type > > > > > > Yes, but if something can be stored JUST as a number, it is not only a > > > storage space issue, but a speed issue. If you took just the 5 digit > > Zip > > > code, (ignoring any extensions), and you wanted to search for >addresses > > in > > > the range of 70000 to 80000, if the numbers are stored as text, the > > > processor is going to have to compare each digit of the number, >because > > > it's > > > looking at ASCII (or unicode) and NOT a number. However, if it's >stored > > > as > > > a number, then it's a VERY simple check to determine if it falls in >that > > > range! > > > > > > Drew _________________________________________________________________ Tired of spam? Get advanced junk mail protection with MSN 8. http://join.msn.com/?page=features/junkmail _______________________________________________ AccessD mailing list AccessD at databaseadvisors.com http://databaseadvisors.com/mailman/listinfo/accessd Website: http://www.databaseadvisors.com