Let me ask the same question as Dossy: Did Informix bring this up in
the public? If so, what is the reference?
Depending on what they actually meant, they might be wrong.
Linux is well capable of guaranteeing successful writes. I.e. when the
data base commits a transaction, it needs to know when the data are
physically on the disk. On Unix, you can take the end of write(2)
for synchronization, provided the system does no write caching. On
BSD they have the raw disk devices for that purpose. On Linux, you
just need to open the device or the file (on ext2) with O_SYNC to
get synchronous writes.
So I think the statement 'You cannot implement a reliable data base
on Linux' is an urban myth.
It is true that Linux has no raw disks in the BSD sense: Linux always
caches data on read, as well as keeping copies in cache on write. Some
people suggest that data bases don't want the cache because they feel
they can manage the free memory better than the OS. Well, you also
can fine-tune the amount of memory the system uses for caching.
Regards,
Martin