All of lore.kernel.org
 help / color / mirror / Atom feed
From: Joseph Qi <joseph.qi@linux.alibaba.com>
To: Heming Zhao <heming.zhao@suse.com>
Cc: ocfs2-devel@lists.linux.dev, ailiop@suse.com
Subject: Re: [PATCH] ocfs2: improve write IO performance when fragmentation is high
Date: Wed, 13 Mar 2024 17:09:53 +0800	[thread overview]
Message-ID: <70d119b5-53c0-4a92-84ec-9b5053184959@linux.alibaba.com> (raw)
In-Reply-To: <ccbfc231-cb82-4455-a137-7d9c9dab44d5@suse.com>



On 3/13/24 9:34 AM, Heming Zhao wrote:
> Hi Joseph,
> 
> Before sending v2 patch, I need to explain ocfs2_find_max_contig_free_bits()
> for you. See the comment in below.
> 
> On 3/11/24 19:04, Joseph Qi wrote:
>>
>> Hi,
>>
>> Please see my comments inline.
>>
>> On 3/8/24 9:52 AM, Heming Zhao wrote:
>>> The group_search function ocfs2_cluster_group_search() should
>>> bypass groups with insufficient space to avoid unnecessary
>>> searches.
>>>
>>> This patch is particularly useful when ocfs2 is handling huge
>>> number small files, and volume fragmentation is very high.
>>> In this case, ocfs2 is busy with looking up available la window
>>> from //global_bitmap.
>>>
>>> This patch introduces a new member in the Group Description (gd)
>>> struct called 'bg_contig_free_bits', representing the max
>>> contigous free bits in this gd. When ocfs2 allocates a new
>>> la window from //global_bitmap, 'bg_contig_free_bits' helps
>>> expedite the search process.
>>>
>>> Let's image below path.
>>>
>>> 1. la state (->local_alloc_state) is set THROTTLED or DISABLED.
>>>
>>> 2. when user delete a large file and trigger
>>>     ocfs2_local_alloc_seen_free_bits set osb->local_alloc_state
>>>     unconditionally.
>>>
>>> 3. a write IOs thread run and trigger the worst performance path
>>>
>>> ```
>>> ocfs2_reserve_clusters_with_limit
>>>   ocfs2_reserve_local_alloc_bits
>>>    ocfs2_local_alloc_slide_window //[1]
>>>     + ocfs2_local_alloc_reserve_for_window //[2]
>>>     + ocfs2_local_alloc_new_window //[3]
>>>        ocfs2_recalc_la_window
>>> ```
>>>
>>> [1]:
>>> will be called when la window bits used up.
>>>
>>> [2]:
>>> under la state is ENABLED, and this func only check global_bitmap
>>> free bits, it will succeed in general.
>>>
>>> [3]:
>>> will use the default la window size to search clusters then fail.
>>> ocfs2_recalc_la_window attempts other la window sizes.
>>> the timing complexity is O(n^4), resulting in a significant time
>>> cost for scanning global bitmap. This leads to a dramatic slowdown
>>> in write I/Os (e.g., user space 'dd').
>>>
>>> i.e.
>>> an ocfs2 partition size: 1.45TB, cluster size: 4KB,
>>> la window default size: 106MB.
>>> The partition is fragmentation by creating & deleting huge mount of
>>> small files.
>>>
>>> before this patch, the timing of [3] should be
>>> (the number got from real world):
>>> - la window size change order (size: MB):
>>>    106, 53, 26.5, 13, 6.5, 3.25, 1.6, 0.8
>>>    only 0.8MB succeed, 0.8MB also triggers la window to disable.
>>>    ocfs2_local_alloc_new_window retries 8 times, first 7 times totally
>>>    runs in worst case.
>>> - group chain number: 242
>>>    ocfs2_claim_suballoc_bits calls for-loop 242 times
>>> - each chain has 49 block group
>>>    ocfs2_search_chain calls while-loop 49 times
>>> - each bg has 32256 blocks
>>>    ocfs2_block_group_find_clear_bits calls while-loop for 32256 bits.
>>>    for ocfs2_find_next_zero_bit uses ffz() to find zero bit, let's use
>>>    (32256/64) (this is not worst value) for timing calucation.
>>>
>>> the loop times: 7*242*49*(32256/64) = 41835024 (~42 million times)
>>>
>>> In the worst case, user space writes 1MB data will trigger 42M scanning
>>> times.
>>>
>>> under this patch, the timing is '7*242*49 = 83006', reduce by three
>>> orders of magnitude.
>>>
>>> Signed-off-by: Heming Zhao <heming.zhao@suse.com>
>>> ---
>>>   fs/ocfs2/move_extents.c |   6 +-
>>>   fs/ocfs2/ocfs2_fs.h     |   4 +-
>>>   fs/ocfs2/resize.c       |   7 +++
>>>   fs/ocfs2/suballoc.c     | 120 ++++++++++++++++++++++++++++------------
>>>   fs/ocfs2/suballoc.h     |  28 +++++++++-
>>>   5 files changed, 122 insertions(+), 43 deletions(-)
>>>
>>> diff --git a/fs/ocfs2/move_extents.c b/fs/ocfs2/move_extents.c
>>> index 1f9ed117e78b..6b9753b8c6fb 100644
>>> --- a/fs/ocfs2/move_extents.c
>>> +++ b/fs/ocfs2/move_extents.c
>>> @@ -576,6 +576,7 @@ static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
>>>       u32 move_max_hop = ocfs2_blocks_to_clusters(inode->i_sb,
>>>                               context->range->me_threshold);
>>>       u64 phys_blkno, new_phys_blkno;
>>> +    struct ocfs2_suballoc_result res = {0};
>>>         phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
>>>   @@ -684,8 +685,9 @@ static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
>>>           goto out_commit;
>>>       }
>>>   -    ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh,
>>> -                     goal_bit, len);
>>> +    res.sr_bit_offset = goal_bit;
>>> +    res.sr_bits = len;
>>> +    ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh, &res, 0);
>>>       if (ret) {
>>>           ocfs2_rollback_alloc_dinode_counts(gb_inode, gb_bh, len,
>>>                              le16_to_cpu(gd->bg_chain));
>>> diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
>>> index 7aebdbf5cc0a..eae18d772e93 100644
>>> --- a/fs/ocfs2/ocfs2_fs.h
>>> +++ b/fs/ocfs2/ocfs2_fs.h
>>> @@ -883,14 +883,14 @@ struct ocfs2_group_desc
>>>       __le16    bg_free_bits_count;     /* Free bits count */
>>>       __le16   bg_chain;               /* What chain I am in. */
>>>   /*10*/    __le32   bg_generation;
>>> -    __le32    bg_reserved1;
>>> +    __le32   bg_contig_free_bits;  /* max contig free bits length */
>>>       __le64   bg_next_group;          /* Next group in my list, in
>>>                          blocks */
>>>   /*20*/    __le64   bg_parent_dinode;       /* dinode which owns me, in
>>>                          blocks */
>>>       __le64   bg_blkno;               /* Offset on disk, in blocks */
>>>   /*30*/    struct ocfs2_block_check bg_check;    /* Error checking */
>>> -    __le64   bg_reserved2;
>>> +    __le64   bg_reserved1;
>>>   /*40*/    union {
>>>           DECLARE_FLEX_ARRAY(__u8, bg_bitmap);
>>>           struct {
>>> diff --git a/fs/ocfs2/resize.c b/fs/ocfs2/resize.c
>>> index d65d43c61857..ab30518d5f04 100644
>>> --- a/fs/ocfs2/resize.c
>>> +++ b/fs/ocfs2/resize.c
>>> @@ -91,6 +91,7 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
>>>       u16 cl_bpc = le16_to_cpu(cl->cl_bpc);
>>>       u16 cl_cpg = le16_to_cpu(cl->cl_cpg);
>>>       u16 old_bg_clusters;
>>> +    u32 contig_bits = 0, old_bg_contig_free_bits;
>>>         trace_ocfs2_update_last_group_and_inode(new_clusters,
>>>                           first_new_cluster);
>>> @@ -122,6 +123,11 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
>>>           le16_add_cpu(&group->bg_free_bits_count, -1 * backups);
>>>       }
>>>   +    ocfs2_find_max_contig_free_bits(group->bg_bitmap,
>>> +                    group->bg_bits, 0, &contig_bits);
>>> +    old_bg_contig_free_bits = group->bg_contig_free_bits;
>>> +    group->bg_contig_free_bits = cpu_to_le32(contig_bits);
>>> +
>>>       ocfs2_journal_dirty(handle, group_bh);
>>>         /* update the inode accordingly. */
>>> @@ -160,6 +166,7 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
>>>           le16_add_cpu(&group->bg_free_bits_count, backups);
>>>           le16_add_cpu(&group->bg_bits, -1 * num_bits);
>>>           le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits);
>>> +        group->bg_contig_free_bits = old_bg_contig_free_bits;
>>>       }
>>>   out:
>>>       if (ret)
>>> diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
>>> index 166c8918c825..df149c3b0eed 100644
>>> --- a/fs/ocfs2/suballoc.c
>>> +++ b/fs/ocfs2/suballoc.c
>>> @@ -37,21 +37,6 @@
>>>     #define OCFS2_MAX_TO_STEAL        1024
>>>   -struct ocfs2_suballoc_result {
>>> -    u64        sr_bg_blkno;    /* The bg we allocated from.  Set
>>> -                       to 0 when a block group is
>>> -                       contiguous. */
>>> -    u64        sr_bg_stable_blkno; /*
>>> -                         * Doesn't change, always
>>> -                         * set to target block
>>> -                         * group descriptor
>>> -                         * block.
>>> -                         */
>>> -    u64        sr_blkno;    /* The first allocated block */
>>> -    unsigned int    sr_bit_offset;    /* The bit in the bg */
>>> -    unsigned int    sr_bits;    /* How many bits we claimed */
>>> -};
>>> -
>>>   static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
>>>   {
>>>       if (res->sr_blkno == 0)
>>> @@ -1272,6 +1257,25 @@ static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
>>>       return ret;
>>>   }
>>>   +/* caller should initialize contig_bits */
>>> +void ocfs2_find_max_contig_free_bits(void *bitmap,
>>> +             unsigned int total_bits, int start,
>>> +             unsigned int *contig_bits)
>>> +{
>>> +    int offset, free_bits;
>>> +
>>> +    while (start < total_bits) {
>>> +        offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start);
>>> +        if (offset == total_bits)
>>> +            break;
>>> +
>>> +        start = ocfs2_find_next_bit(bitmap, total_bits, offset);
>>> +        free_bits = start - offset;
>>> +        if (*contig_bits < free_bits)
>>> +            *contig_bits = free_bits;
>>> +    }
>>
>> Or we can just return the contig_bits. Something like:
>> unsigned int ocfs2_find_max_contig_free_bits(void *bitmap,
>>         unsigned int total_bits, int start)
>> {
>>     int offset, start;
>>     unsigned int contig_bits = 0;
>>
>>     while () {
>>         ...
>>     }
>>
>>     return contig_bits;
>> }
> 
> the input *contig_bits has two jobs:
> - return the max contigous free bits of bitmap
> - if *contig_bits is not ZERO, it keeps the max-contig-bits
>   from last search. When the input 'start' not zero
>   (means start from middle of bitmap), the max-contig-bits
>   from latter part of bitmap may smaller than first part.
>   at this case, function should keep & return the input
>   *contig_bits.
> 

IC, I don't think it's a good idea to mix up the 2 responsibilities.
You've hidden the logic that it may change the input contig_bits (not
the case of first search).
You can also implement it something like:

contig_bits = ocfs2_find_max_contig_free_bits();
if (config_bits > sr_max_contig_bits)
	sr_max_contig_bits = contig_bits;

This will be more explicit for an optimized search.

Thanks,
Joseph

> the view of ocfs2_find_max_contig_free_bits():
> 
>   front part  |  latter part
> [-------------+------------------] bitmap
>  0            'start'           total_bits
> 
> the call stack:
> 
> ocfs2_search_one_group
> + ac->ac_group_search
> |  ocfs2_cluster_group_search
> |   ocfs2_block_group_find_clear_bits
> |    + scan area [0, res->sr_bit_offset + res->sr_bits - 1]
> |    | /* zhm: save the second max-contig-bits.
> |    |  * if it doesn't exist, the value is 0. */
> |    + res->sr_max_contig_bits = prev_best_size;
> |
> + ocfs2_block_group_set_bits
>    ocfs2_find_max_contig_free_bits(...,
>             start, /* zhm: may not ZERO, help to speed up the search */
>             res->sr_max_contig_bits) /* zhm: the optimal value may have been reached */
> 
> So we should keep ocfs2_find_max_contig_free_bits() with a return type of 'void'.
>

  reply	other threads:[~2024-03-13  9:10 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-03-08  1:52 [PATCH] ocfs2: improve write IO performance when fragmentation is high Heming Zhao
2024-03-11  9:18 ` Joseph Qi
2024-03-11 11:04 ` Joseph Qi
2024-03-12  0:54   ` Heming Zhao
2024-03-13  1:34   ` Heming Zhao
2024-03-13  9:09     ` Joseph Qi [this message]
2024-03-13 12:32       ` Heming Zhao
2024-03-19  6:57 Heming Zhao
2024-03-19  7:00 ` Heming Zhao

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=70d119b5-53c0-4a92-84ec-9b5053184959@linux.alibaba.com \
    --to=joseph.qi@linux.alibaba.com \
    --cc=ailiop@suse.com \
    --cc=heming.zhao@suse.com \
    --cc=ocfs2-devel@lists.linux.dev \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.