you'd have to look far and wide for a filesystem that still imposes that lock on writes, WHEN direct IO is used... VxFS,PolyServe, Sol UFS (forcedirectio mount) and most other legacy unix derivations pulled that lock for directio use cases long, long ago. Mid 90's. Heck, I remember measuring the goodness of that fix on Sequent Dynix with Oracle 6.0.27 back in 1990 :-) ________________________________ From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Hameed, Amir Sent: Friday, September 02, 2005 4:48 PM To: oracle-l@xxxxxxxxxxxxx Subject: RE: Choosing data file size for a multi TB database? On very large data files running on buffered filesystem, wouldn't the single-writer lock cause foreground processes (that are trying to read data) to wait when the DBWR is check pointing? ________________________________ From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Branimir Petrovic Sent: Friday, September 02, 2005 6:54 PM To: oracle-l@xxxxxxxxxxxxx Subject: RE: Choosing data file size for a multi TB database? What about checkpoint against tens of thousands of data files, surely more-merrier rule holds? For that reason (or due to a fear factor) I was under may be false impression that smaller number (in hundreds) of relatively larger data files (20 GB or so) might be better choice. Other very real problem with 10TB database I can easily foresee, but for which I do not know proper solution, is how would one go about the business of regular verification of taped backup sets? Have another humongous hardware just for that purpose? Fully trust the rust? (i.e. examine backup logs and never try restoring, or...) What do people do to ensure multi TB monster databases are surely and truly safe and restorable/rebuildable? Branimir -----Original Message----- From: Tim Gorman [mailto:tim@xxxxxxxxx] Sent: Friday, September 02, 2005 5:59 PM To: oracle-l@xxxxxxxxxxxxx Subject: Re: Choosing data file size for a multi TB database? Datafile sizing has the greatest regular impact on backups and restores. Given a large multi-processor server with 16 tape drives available, which would do a full backup or full restore fastest? * a 10-Tbyte database comprised of two 5-Tbyte datafiles * a 10-Tbyte database comprised of ten 1-Tbyte datafiles * a 10-Tbyte database comprised of two-hundred 50-Gbyte datafiles? * a 10-Tbyte database comprised of two-thousand 5-Gbyte datafiles? Be sure to consider what type of backup media are you using, how much concurrency will you be using, and the throughput of each device? There is nothing "unmanageable" about hundreds or thousands of datafiles; don't know why that's cited as a concern. Oracle8.0 and above has a limitation on 65,535 datafiles per tablespace, but otherwise large numbers of files are not something to be concerned about. Heck, the average distribution of a Java-based application is comprised of 42 million directories and files and nobody ever worries about it...