[Pvfs2-developers] Patches: (1) fixing incorrect small I/O total_size, (2) processing multiple aiocbs at a time

[Rob Ross]

good catch on the aiocb stuff avery; that one has been in there for 
years i would guess.

rob

Avery Ching wrote:
> On Thu, 2006-05-18 at 17:26 -0500, Robert Latham wrote:
>> Thanks for the patches.  tracking down these bugs can be a big pain.
>> good job, man.  just a few questions.
>>
>> On Thu, May 18, 2006 at 03:44:28PM -0600, Avery Ching wrote:
>>> small_io_total_bytes_fix.patch:
>>>
>>> Fixed an issue with the small I/O total size being passed to the
>>> trove_bstream_write_list().  The uint32_t was being casted to 
>>> (TROVE_size *)&s_op->req->u.small_io.total_bytes on line 162 of
>>> small_io.sm.
>> If you are going to change the type of 'total_bytes', don't forget
>> you'll have to change the encode_PVFS_servreq_small_io and
>> decode_PVFS_servreq_small_io macros, too.  
>>
> 
> You're right that this might affect other stuff.  I think I guess the
> encode and decode functions will need to be changed.  Okay, fixed.  I've
> attached another patch that does this and changes it to PVFS_size.  I
> think that's it hopefully...
> 
>>> process_multiple_aiocbs.patch:
>>>
>>> Fixed the dbpf_bstream_listio_convert() function to actually handle more
>>> than a single aiocb at a time.  This should boost noncontiguous I/O
>>> performance quite a bit.
>> Can you help me understand why your changes work?  What is making the
>> existing code return too early?   I'm not saying you did it wrong, of
>> course.  it's just code i don't know very well.
>>
> 
> In most cases (actually all that I know of) mem_count = 1.  This means
> that when it passes through the while loop it breaks out even if it
> isn't finished since it is oom.  Therefore it was only doing a single
> aiocb at a time.  By changing the break conditional and updating the
> cur_mem_size and cur_stream_size, we can see where we really need to
> break out of the while loop (when it is oos or oom AND there are no
> bytes left in the current piece).  Hope that helps!
> 
>> ==rob	
>>
>>
>> ------------------------------------------------------------------------
>>
>> ? module.mk
>> Index: pvfs2-req-proto.h
>> ===================================================================
>> RCS file: /anoncvs/pvfs2/src/proto/pvfs2-req-proto.h,v
>> retrieving revision 1.138
>> diff -u -r1.138 pvfs2-req-proto.h
>> --- pvfs2-req-proto.h	24 Jan 2006 21:33:24 -0000	1.138
>> +++ pvfs2-req-proto.h	18 May 2006 23:03:38 -0000
>> @@ -953,7 +953,7 @@
>>      PVFS_offset offsets[SMALL_IO_MAX_SEGMENTS];
>>      PVFS_size sizes[SMALL_IO_MAX_SEGMENTS];
>>  
>> -    uint32_t total_bytes;
>> +    PVFS_size total_bytes; /* changed from int32_t */
>>      void * buffer;
>>  };
>>  
>> @@ -968,7 +968,7 @@
>>      encode_PINT_Request(pptr, &(x)->file_req); \
>>      encode_PVFS_offset(pptr, &(x)->file_req_offset); \
>>      encode_PVFS_size(pptr, &(x)->aggregate_size); \
>> -    encode_uint32_t(pptr, &(x)->total_bytes); \
>> +    encode_PVFS_size(pptr, &(x)->total_bytes); \
>>      encode_skip4(pptr,); \
>>      if ((x)->io_type == PVFS_IO_WRITE) \
>>      { \
>> @@ -994,7 +994,7 @@
>>      PINT_request_decode((x)->file_req); /* unpacks the pointers */ \
>>      decode_PVFS_offset(pptr, &(x)->file_req_offset); \
>>      decode_PVFS_size(pptr, &(x)->aggregate_size); \
>> -    decode_uint32_t(pptr, &(x)->total_bytes); \
>> +    decode_PVFS_size(pptr, &(x)->total_bytes); \
>>      decode_skip4(pptr,); \
>>      if ((x)->io_type == PVFS_IO_WRITE) \
>>      { \
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Pvfs2-developers mailing list
>> Pvfs2-developers at beowulf-underground.org
>> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers