RE: raid0 and iscsi

[jshankar <jshankar@CS.ColoState.EDU>]

Hello Neil,


Thanks for the reply.

I was working on the internet scsi(iscsi drivers provided by intel) and 
software
raid configuration for my thesis in Storage Area Network.

The problems right now i am facing is
1> For a local disk write  for 512 mb in DELL 1.2 Ghz it takes me 1 min.
2> For a single iscsi device write, the transfer of data takes place at
4000 bytes/sec.
2> For raid configuration, the rate is pretty slow ( 40 bytes/sec).
One thing i have observe is that write to one of the device( from a total of 2 
device)  takes for few minutes and then stop.


I was trying to figure what routines in the operating system code, I might 
need
to look into to understand a problem if there is one.

 Some more problems.
If one of the network comes down, the mkfs for the raid , hangs.


>A raid0 array can be made of a number of drives of differing sizes.
>To accomodate this we divide the address space into several blocks.
>The first block is striped across all drives to the size of the
>smallest.  The next block is striped across the remaining drives to
>the size of the next smallest, etc.

Is there a relation between chunk size and block size of raid. Is this similar 
to buffer size( write(fd,&buffer,buffer_size) and block size of filesystem.

-Jay




>===== Original Message From Neil Brown <neilb@cse.unsw.edu.au> =====
>On Sunday October 12, jshankar@CS.ColoState.EDU wrote:
>> Hello,
>>
>>
>> I was testing the raid0 configuration for the iscsi device ( /dev/sdb and
>> /dev/sdc).
>>
>> For writing a 512MB in a LAN environment it was taking 24 hrs. i was going
>> through the source code of raid0.c and certain things doesn't made sense to 
me.
>>
>> 1> what role does the Hash bit play.
>
>A raid0 array can be made of a number of drives of differing sizes.
>To accomodate this we divide the address space into several blocks.
>The first block is striped across all drives to the size of the
>smallest.  The next block is striped across the remaining drives to
>the size of the next smallest, etc.
>
>Mapping from a virtual device address to block and thence a drive and
>offset is not straight forward.  It requires a table search.  The hash
>table helps accelerate this search.
>
>
>> 2>If my chunk size is 8 byte. Does that mean it will write 8 byte into 
device 0
>> and then into device 1. Is the write request to the disk in synchronous or
>> asynchronous mode.
>>
>You cannot have an 8byte chunk size.  4K is the minimum.
>With an 8K chunk size and 2 devices,
> sectors between 0 and 8K, 16k and 24K, 32K and 40K, 48K and 56K etc
>    are written to the first device.
> sectors between 8K and 16K, 28K and 32K, 40K and 48K etc
>    are written to the second device.
>
>raid0 does not impose any synchronisation.  Writes are only
>synchronous if the filesystem waits for them.  raid0 never waits.
>
>>
>> 3> Is wite_disk_sb in md.c responsible for writing into disk??. Is 
fsync_dev
>> responsible for synhronous write ??. If so can i change to asynchronous 
write.I
>> will really appreciate if somebody can tell me what all routines I need to 
go
>> through to figure out the functionality of raid0 behaviour.
>
>write_disk_sb is for writing the raid superblock to disk.  It doesn't
>happen often.
>fsync_dev is fairly irrelevant. You can safely ignore it.
>
>I hope that helps.
>
>NeilBrown
>
>>
>> Thanks
>> Jayshankar
>>
>> -
>> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html