https://jira.whamcloud.com This file is an XML representation of an issue en-us 9.4.14 940014 05-12-2023 https://jira.whamcloud.com/browse/LU-2565 Lustre <p>I noticed on the server that the ldiskfs_inode_cache is <em>just</em> over 1kB in size (1048 bytes on 2.7.56), which means only 3 inodes can fit into a 4kB page. It would be useful to reduce the inode size to be &lt;= 1024 bytes in order to improve the memory efficiency by 1/3.</p> LU-2565

reduce ldiskfs inode size

Improvement Minor Resolved Low Priority WC Triage Andreas Dilger easy Thu, 3 Jan 2013 14:14:05 +0000 Wed, 13 Feb 2019 07:12:20 +0000 Wed, 13 Feb 2019 07:12:20 +0000 Lustre 2.5.0 Lustre 2.8.0 0 3 <p>On RHEL 6.6 the <tt>ldiskfs_inode_info</tt> is 1056 bytes in size, but is only using 1040 bytes, so it could be shrunk with a bit of effort to reduce MDS memory usage by 1/3 for this slab. This can be seen easily with the <tt>pahole</tt> tool from the <tt>dwarves</tt> package:</p> <div class="code panel" style="border-width: 1px;"><div class="codeContent panelContent"> <pre class="code-java"># pahole -C ldiskfs_inode_info ldiskfs/ldiskfs/ldiskfs.ko | struct ldiskfs_inode_info { __le32 i_data[15]; <span class="code-comment">/* 0 60 */</span> __u32 i_dtime; <span class="code-comment">/* 60 4 */</span> <span class="code-comment">/* --- cacheline 1 boundary (64 bytes) --- */</span> ldiskfs_fsblk_t i_file_acl; <span class="code-comment">/* 64 8 */</span> struct semaphore i_append_sem; <span class="code-comment">/* 72 24 */</span> ldiskfs_lblk_t i_dir_start_lookup; <span class="code-comment">/* 96 4 */</span> ldiskfs_group_t i_block_group; <span class="code-comment">/* 100 4 */</span> <span class="code-object">long</span> unsigned <span class="code-object">int</span> i_state_flags; <span class="code-comment">/* 104 8 */</span> <span class="code-object">long</span> unsigned <span class="code-object">int</span> i_flags; <span class="code-comment">/* 112 8 */</span> struct rw_semaphore xattr_sem; <span class="code-comment">/* 120 32 */</span> <span class="code-comment">/* --- cacheline 2 boundary (128 bytes) was 24 bytes ago --- */</span> struct list_head i_orphan; <span class="code-comment">/* 152 16 */</span> loff_t i_disksize; <span class="code-comment">/* 168 8 */</span> struct rw_semaphore i_data_sem; <span class="code-comment">/* 176 32 */</span> <span class="code-comment">/* --- cacheline 3 boundary (192 bytes) was 16 bytes ago --- */</span> struct inode vfs_inode; <span class="code-comment">/* 208 592 */</span> <span class="code-comment">/* --- cacheline 12 boundary (768 bytes) was 32 bytes ago --- */</span> struct jbd2_inode jinode; <span class="code-comment">/* 800 48 */</span> <span class="code-comment">/* --- cacheline 13 boundary (832 bytes) was 16 bytes ago --- */</span> <span class="code-object">long</span> unsigned <span class="code-object">int</span> i_ext_generation; <span class="code-comment">/* 848 8 */</span> struct ldiskfs_ext_cache i_cached_extent; <span class="code-comment">/* 856 24 */</span> <span class="code-comment">/* XXX last struct has 4 bytes of padding */</span> struct timespec i_crtime; <span class="code-comment">/* 880 16 */</span> <span class="code-comment">/* --- cacheline 14 boundary (896 bytes) --- */</span> struct list_head i_prealloc_list; <span class="code-comment">/* 896 16 */</span> spinlock_t i_prealloc_lock; <span class="code-comment">/* 912 4 */</span> ldiskfs_group_t i_last_alloc_group; <span class="code-comment">/* 916 4 */</span> unsigned <span class="code-object">int</span> i_reserved_data_blocks; <span class="code-comment">/* 920 4 */</span> unsigned <span class="code-object">int</span> i_reserved_meta_blocks; <span class="code-comment">/* 924 4 */</span> unsigned <span class="code-object">int</span> i_allocated_meta_blocks; <span class="code-comment">/* 928 4 */</span> <span class="code-comment">/* XXX 4 bytes hole, <span class="code-keyword">try</span> to pack */</span> sector_t i_da_metadata_calc_last_lblock; <span class="code-comment">/* 936 8 */</span> <span class="code-object">int</span> i_da_metadata_calc_len; <span class="code-comment">/* 944 4 */</span> <span class="code-object">short</span> unsigned <span class="code-object">int</span> i_delalloc_reserved_flag; <span class="code-comment">/* 948 2 */</span> __u16 i_extra_isize; <span class="code-comment">/* 950 2 */</span> spinlock_t i_block_reservation_lock; <span class="code-comment">/* 952 4 */</span> <span class="code-comment">/* XXX 4 bytes hole, <span class="code-keyword">try</span> to pack */</span> <span class="code-comment">/* --- cacheline 15 boundary (960 bytes) --- */</span> qsize_t i_reserved_quota; <span class="code-comment">/* 960 8 */</span> struct list_head i_aio_dio_complete_list; <span class="code-comment">/* 968 16 */</span> spinlock_t i_completed_io_lock; <span class="code-comment">/* 984 4 */</span> <span class="code-comment">/* XXX 4 bytes hole, <span class="code-keyword">try</span> to pack */</span> ldiskfs_io_end_t * cur_aio_dio; <span class="code-comment">/* 992 8 */</span> atomic_t i_aiodio_unwritten; <span class="code-comment">/* 1000 4 */</span> tid_t i_sync_tid; <span class="code-comment">/* 1004 4 */</span> struct mutex i_aio_mutex; <span class="code-comment">/* 1008 32 */</span> <span class="code-comment">/* --- cacheline 16 boundary (1024 bytes) was 16 bytes ago --- */</span> __u64 i_fs_version; <span class="code-comment">/* 1040 8 */</span> tid_t i_datasync_tid; <span class="code-comment">/* 1048 4 */</span> <span class="code-comment">/* size: 1056, cachelines: 17, members: 37 */</span> <span class="code-comment">/* sum members: 1040, holes: 3, sum holes: 12 */</span> <span class="code-comment">/* padding: 4 */</span> <span class="code-comment">/* paddings: 1, sum paddings: 4 */</span> <span class="code-comment">/* last cacheline: 32 bytes */</span> }; </pre> </div></div> <p><tt>ldiskfs_inode_info</tt> is up to 1072 bytes in size with 1050 bytes in use, so shrinking it seems difficult at this point.</p> Related LU-1548 Development Rank 1|hzveiv: Rank (Obsolete) 5994