Re: iostat , sun ufs io rate
Posted in 2005
Topics: General Discussion
Thanks Paul , For the other question ,can you make a comment based on your experiences? When i issue a mkfile command in order to create a file , i can see more or less 1mb. write or read per read operation. This is what i configured for ufs file system. But when i issue an oracle query , the kb. read or write per io operation is very low. ( For the oracle part , i am also monitoring what the dbserver do during this query , oracle issues 1mb. read requests to the file system, and since i use parallel query which bypasses the oracle buffer cache i am sure that oracle requests the data from the disk subsystem .) Kind Regards, hope
hopehope_123 wrote: > Thanks Paul , > > For the other question ,can you make a comment based on your > experiences? When i issue a mkfile command in order to create a file , > i can see more or less 1mb. write or read per read operation. This is > what i configured for ufs file system. But when i issue an oracle query > , the kb. read or write per io operation is very low. ( For the oracle > part , i am also monitoring what the dbserver do during this query , > oracle issues 1mb. read requests to the file system, and since i use > parallel query which bypasses the oracle buffer cache i am sure that > oracle requests the data from the disk subsystem .) > > > Kind Regards, > hope > Can't comment I've been fortunate enough never have had to use Oracle -- Paul Watson # Oninit Ltd # Growing old is mandatory Tel: +44 1436 672201 # Growing up is optional Fax: +44 1436 678693 # Mob: +44 7818 003457 # www.oninit.com #
truss the oracle process...see what it is really doing truss -aef -v all -o outfile -p <pid> Perhaps Oracle is using an index and so as it goes down the index the pages it needs are not contiguous so it does random reads rather than the sequential i/o that mkfile does. That will drop performance but cannot be helped. With a filesystem you are going thought the filessystem code hence reading inodes, single, double and triple indirect pointers and having to push all that data through the cpu. More overhead... Use raw disk instead!