Messages in this thread | | | Date | Thu, 30 Mar 2000 14:42:43 +0400 | From | Hans Reiser <> | Subject | Re: Is ReiserFS really a journaling file system, or is it really just a synchronous-metadata file system like BSD FFS? |
| |
In the long term I see this as the right approach for us, we just need to gain some programmer bandwidth and it can happen.
Hans
Toon van der Pas wrote: > > On Wed, Mar 29, 2000 at 09:08:32AM +0200, Marc SCHAEFER wrote: > > Scott Henry <scotth@sgi.com> wrote: > > > You are thinking of the other type, a log-based filesystem, where > > > metadata and user data are logged. It is much more difficult to > > > > Maybe I am mixing up oranges and logs, but for me, in certain cases, > > a log-based filesystem can be way faster than a normal or > > metadata-log filesystem: as data is only appended, you can do > > the append sequentially and in very big chunks (say, 4 MBytes > > sequential I/Os). You will get very good performance from > > single device and hardware (or even software) RAID0. And > > you don't need to think about seek times when writing. > > > > Of course, in theory this also means you have to add a new disk > > drive every now and then, unless some data reclaimation is > > done. Which brings in new problems. > > > > This can of log file system (with data reclaimation) ressembles > > a bit the HSM system when using tapes for data migration. > > The way this is done on Tandem NonStop systems requires cooperation > of the applications: > > 1. Files/tables that need to be logged are marked with an AUDIT flag. > > 2. Applications define complete transactions by means of calls like > BEGINTRANSACTION, ENDTRANSACTION, ABORTTRANSACTION. > When a process fails, active transaction(s) associated with that > process are rolled back automatically by the Transaction Monitoring > Facility. > > 3. The transactions are logged sequentially to socalled 'audittrails' > on separate disks. > > 4. Regularly 'online dumps' are made of the audited tables. > File recovery is possible by restoring from the online dump and > replaying the audittrails. The consistency of the database is > garanteed. > > 5. Audittrails are backed up to cheaper media and subsequently released, > once they do not contain active transactions any longer. > > 6. Online dumps and backed up audittrails are automatically released, > once they are no longer needed for file/volume recovery. > (i.e. there are newer online dumps available) > > On Tandem you'll get much better performance _with_ auditing (logging) > transactions than without. > > Toon. > > - > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in > the body of a message to majordomo@vger.rutgers.edu > Please read the FAQ at http://www.tux.org/lkml/
-- You can get ReiserFS at http://devlinux.org/namesys, and customizations and industrial grade support at reiser@idiom.com.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |