Re: A few questions

>A few of questions: No kidding! :-) [WARNING: I haven't really woken up yet! :-] >--- >In 7.3 I see that after backing up old log files, they get marked U and B. >In 5.0, I see that instead of marking the logs U (used), and B (backed-up), it marks them as F (free). >Is this behavior normal, or is something wrong with my 7.3 instance. >I am just wondering when those logs marked as U and B become F (free). This is normal, all version 7 ports (and probably 6 as well) behave like this. It's a bit irritating, because some of the course notes and (I think) some of the documentation still refers to the old "F" behaviour. So your v7 logs will never be marked as free. >--- >There is no USERS parameter in the onconfig file for 7.3. >Does this mean that 7.3 has hardcoded limits for the # of users? Not really, it just means that v7 will manage its users in a different way, via the NETTYPE settings. You define how many communication threads you want and how many users can connect to each. This (more or less, but not quite because of system overheads) gives you the number of users that can connect. >--- >I read a few posts a while back about the possibility of the optimizer being too slow to determine a query path. >I believe it was on a 7.3X instance. >My main question is, I myself have dissapointed by the raw speed of 7.3 vs 5.0, and is this a likely place to look for possible performance improvements? >If so, is there anyway within 4gl RDS to code apps to minimize the length of time the optimizer evaluates its query plan possibilities? You have to do a *lot* more tuning to get 7 to match 5's performance, especially on relative small or low-user-count systems. Particular things to investigate are: 1. SET OPTIMISATION LOW 2. Subquery flattening (search dejanews for previous discussions) 3. OPTCOMPIND=0 in $ONCONFIG 4. The performance tuning course. :-) >--- >If I want to split high use tables across drives, what are some things that I should consider when deciding what goes where. It depends. If you're trying to improve bulk load performance, use ROUND ROBIN; if you're doing equality comparisons (something = something else) and you don't know the data distributions, use a MOD expression (on the "something" field) if possible; if you know the data distribution and you want to do range or equality comparisons, use a set of expressions based on the "something" field(s). If you do lots of things and don't konw the data distribution because it changes regularly, *try* ROUND ROBIN. You should at least get some parallelism that way. In an OLTP system, you should always try to fragment by expression, however. NOTE: *Always* detach your indexes to a separate dbspace *especially* if you're doing a ROUND ROBIN fragment. Fragmentation doesn't always help, I'm afraid you have to try a couple of things in your environment and see what happens. The above are just *very* broad thoughts on the matter. HTH. ______________________________________________________ Get Your Private, Free Email at http://www.hotmail.com