Messages in this thread | | | Date | Fri, 05 Nov 1999 10:16:06 -0700 | From | "Jeff V. Merkey" <> | Subject | Re: Linux Buffer Cache Does Not Support Mirroring |
| |
Pavel,
Please let me know what you think is the best method for this. I have forwarded the code to you to look at so you know what is needed.
Jeff
"Jeff V. Merkey" wrote: > > OK. > > So now I am convinced. I will need some help with the md driver for > netware mirroring. perhaps it makes the most sense to merge the current > raid1 driver or simply enhance it's capabilities to support what we > need. Here's what we need, tell me if these things are possible with > your current code. > > 1. must support block (4K block) IO r/w. Must also support moving > "window" so the cache above can still get access to blocks during > remirroring. > > 2. mirroring support is very easy. remirroing is very hard. does your > driver support remirroring? our coe does and handles multiplr access > cases. > > 3. Must be able to call into cache to locate data if a read failure > occurs on the device. In the netware mode, during read errors, the > mirroring driver first attempts to read the failing sectors (in groups > of 8 - 4K) by changing the read order (interleave) since this sometimes > will get all the data. If this fails then it queries the buffer cache > for the missing block, it the block is not found, then it asks one of > the mirrors (if present and in sync) for a copy prior to read > hotfixing. > > hotfixing is a reserved area on the partition that is used if any > sectors go bad on the partition. In such a case, the md driver will > need to remap (transparently) the failing sectors to a hotfix area. We > keep tables at offsets 20, 40, 60, and 80 of the hitfix and mirror table > descriptors, and the hotfix table itself immediately follows these > descriptors. > > We could attempt to support what we need in am md driver, but these > drivers are all based on physical LBA rather than logical 4K block, > which is how all these tables are maintained. What's there is somewhat > inflexible and hard to map. > > The mirroring/hotfix code is attached. > > Please advise, > > Jeff > > Pavel Machek wrote: > > > > Hi! > > > > > Most of the code for the md drivers lacks documentation. However, RAID1 > > > is disk based, or am I missing something? THis is what I meant. > > > > I've ran raid5 over local partition and 2 files mounted using network > > block device. And yes, RAID1 is partition based, so you probably are > > missing something. With network block device, you can even do raid1 > > over one local partition and one remote partition, which is excelent > > for high availability. > > > > Pavel > > -- > > The best software in life is free (not shareware)! Pavel > > GCM d? s-: !g p?:+ au- a--@ w+ v- C++@ UL+++ L++ N++ E++ W--- M- Y- R+ > > ------------------------------------------------------------------------ > Name: nwvp.c > nwvp.c Type: Notepad File (application/x-unknown-content-type-cfile) > Encoding: base64
- 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/
| |