>>>>> "Tim" == Tim Bunce <Tim.Bunce@stripped> writes:
Tim> On Wed, Jan 12, 2000 at 12:55:02PM +0200, Tonu Samuel wrote:
>> Only striping is implemented yet and there is no concrete plans for
>> future. At moment we hadprimary goal to break 2GB file limit and
>> secondary goal to get more speed. More speed can be achieved if these
>> directories with numbers are symlinked or mounted to different physical
>> disks. This should increase speed in heavy IO conditions.
Tim> Are you relying on the operating system + separate disk drives etc to
Tim> achieve that or will there be separate threads handling the I/O for each?
>> RAID_CHUNKS means that we have 4 directories
Tim> Using the word CHUNKS for that seems wrong, at least for the current
Tim> behaviour. Those aren't chunks, their things that hold many chunks.
>> and RAID_CHUNKSIZE means data is striped in chunks of 16kbytes.
Tim> Having RAID_CHUNKSIZE be specified in 1Kbyte units would probably make
Tim> life easier for people: RAID_CHUNKSIZE=16 (or RAID_CHUNKSIZE=16K).
Good point; We will change that.
Tim> Adding a RAID_FIRSTCHUNK=N (or RAID_OFFSET=N, or RAID_FIRST=N) would be
Tim> a simple way to allow greater control over spreading the I/O.
Tim> Currently there's no way of adding a table that won't add I/O to low
Tim> numbered 'chunks'. For example, I might want to have one table in ./00/
Tim> and ./01/ and another table in ./02/ and ./03/.
This is a bit harder to do as one in this case has to change the MyISAM
interface once again, but it shouldn't be impossible to do.
Tim> Also, what are the implications for crash recovery?
This is done on the MyISAM level so myisamchk will work on the hunks
as if they would one file. In other words; Normally it should work
just fine :)