(Illustration by Gaich Muramatsu)
> > Is there an official limit on RVM log and metadata file sizes? I set my > > log size to 2GB in order to be able to handle files up to 2GB. Ideally, I > > wanted to be able to handle a FS up to 0.5TB, so I wanted to make a > > metadata file of about 16GB. The biggest I've managed to get to work is > > 4GB, otherwise things fail with an error. Does this sound about right? > > ISTR that the protocol for communicating among Coda hosts was > 32-bit-size-oriented, but recently may have been updated to 64-bit > sizes at the cost of backward compatibility, so it might be on a > branch. Anyway, this would impose a hard limit of 4GB (2GB if signed) > on file sizes. Note that this is a wire protocol issue, so a global > substitution of off64_t for int won't help, but rather just introduce > bugs. Sure, I understand that, but if that is a wire protocol issue, then presumably, if I use the same version of Coda (e.g. 6.9.3) on both the IA32 and the x86-64 machine, they will still communicate correctly. Or am I misunderstanding? > > On a related note, are any of these limits (file/log/metadata) > > architecture dependant? If so are there any issues with having an > > x86-64 and an IA32 node in the same replicated cluster? > > I don't think they're CPU-dependent, rather they depend on (a) large > file support in the OS, and (b) a change in the Coda wire protocol. Right - so if both Coda versions are the same and the same kernel version is on both machines it should work? GordanReceived on 2008-01-30 16:20:47