RE: The old raw devices chestnut.
Posted in 2004
Andrew, The question is not UNIX partition header itself, the question is how Oracle places data with respect to the beginning if the raw device. With Informix, You specify manually the 'device offset' with '-o' 'Onspaces' flag. Oracle makes this offset automatically for You if You place Your data on raw device; it doesn't make this offset if You place You data onto a cooked file. As a result, You might get into trouble if You try to move Oracle data (or restore from backup, which is almost the same with Oracle) from cooked file onto raw devices. The problem is that You must copy the data with offset (which is platform-specific) (like: dd if=myfile of=/dev/mydevice obs=10240 seek=50) With Informix, it is more simple to move from files to raw devices - if particular *ancient* platform doesn't require 'offset'. This is just an example of REAL ADMINISTRATIVE OVERHEAD related to raw devices :-) - luckily, not with Informix... Alexey Sonkin -----Original Message----- From: Andrew Hamm [mailto:ahamm@mail.com] Sent: Wednesday, April 14, 2004 9:21 PM To: informix-list@iiug.org Subject: Re: The old raw devices chestnut. Alexey Sonkin wrote: > Oracle uses some offset when it placed data onto raw devices. When You > restore data previously kept on cooked files onto raw devices, You > need to copy files using this offset (smth like 500 kBytes on Digital > UNIX, may very on other platforms) I think you might be pointing to a bit of ancient history (I hope it's ancient history on all modern UNIX) Time was, the partition table, or information about each partition was stored in the first few sectors of the disk or partition on some UNIX. This was not an issue with filesystems of course, because the local O/S utilities took care of building them in just the right place with just the right size. I can't think of any contempory UNIX (that I've used recently;-) which does something this crude, although if you try to attach your raw spaces to the very low level device names (the device names which address actual ferrite particles) then you would probably run into this issue. Using names from the logical volume manager is definitely the way to go. sending to informix-list