Re: [GENERAL] problems with memory
От | selkovjr@mcs.anl.gov |
---|---|
Тема | Re: [GENERAL] problems with memory |
Дата | |
Msg-id | 200002190008.SAA11095@mail.xnet.com обсуждение исходный текст |
Ответ на | Re: [GENERAL] problems with memory (Alfred Perlstein <bright@wintelcom.net>) |
Список | pgsql-general |
> * Dean Browett <dbrowett@bizonline.net> [000218 15:55] wrote: > > Hi, > > > > We are running postgres-6.5.3 on a dual pentium 300 machine, 0.5Gb RAM under > > Linux Redhat 6.0 (kernel 2.2.14). The machine we are using sits on a 100Mb > > network and the nics are 3com3c590's. We are also using a DPT Raid > > controller in a raid5 configuration set up as 1 logical drive. > > > > We are try to insert a large amount of data into the database. What happens > > is that when we first start loading data everything is fine. Over a period > > of time (1.5hrs) there is a marked decrease in performance in terms of both > > memory and cpu usage. At this time cpu usage has crept up to 45-50% and > > memory usage is 100Mb and rising slowly and there is only one connection to > > the database. > > > > All the statistics are gleaned from using 'top'. > > > > > > Consequently, the database gets slower and slower until it loads at the rate > > of <3KBps at which point it becomes more exciting to watch paint dry 8-)). > > > > Postgres is the only program running (except for normal system programs) > > that uses a significant amount of memory. > > > > Is Postgres known to leak memory? What causes the high cpu usage? > > > > Any ideas would be greatly appreciated. > > You really haven't given very much information on the rules and constraints > in your tables, one problem that I had was that a constraint on a table > of mine caused extreme slowdown because each row inserted needed to be > validated through a constraint, as the table grew the amount of data that > needed to be scanned for each insert grew exponentially. > > hope this helps, > -Alfred The same is true of any type of index. Make sure you don't have indices defined unitl after the insert. Also, do a COPY instead of INSERT if possible. --Gene
В списке pgsql-general по дате отправления: