summaryrefslogtreecommitdiffstats
path: root/xlators/performance
Commit message (Collapse)AuthorAgeFilesLines
* performance/quick-read: make use of nanosecond resolution of mtime to decide ↵Raghavendra G2010-04-081-2/+6
| | | | | | | | | | whether to keep cache or not. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 801 (Direct io-mode support and related changes in caching translators.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=801
* performance/io-cache: make use of nano second resolution of mtime during ↵Raghavendra G2010-04-083-6/+16
| | | | | | | | | | cache validation. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 801 (Direct io-mode support and related changes in caching translators.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=801
* performance/quick-read: read directly from backend for fds opened with ↵Raghavendra G2010-04-082-2/+12
| | | | | | | | | | O_DIRECT flag. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 723 (enhancements to quick read) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=723
* performance/write-behind: Resume all the consecutive non-write operations in ↵Raghavendra G2010-03-291-3/+0
| | | | | | | | | | the request queue in a single go. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 740 (read-ahead does not work to its full potential when loaded on top of write-behind) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=740
* performance/io-threads: Support for Solaris.Vijay Bellur2010-03-031-0/+3
| | | | | | | | Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 621 (3.0.2 GlusterFS fails on Solaris 10) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=621
* performance/write-behind: fix data corruption while aggregating the adjacent ↵Raghavendra G2010-02-221-4/+55
| | | | | | | | | | | | | | | | | contiguous iobufs into a single iobuf - while aggregating, we should make sure that the destination has enough memory. __wb_collapse_write_bufs assumed that destination vector's iov_base was aligned to the start of an iobuf and hence memory of page_size is available for aggregation. This assumption is not always true, like in the configuration consisting afr->write->io-cache (afr is on top). Refer to the bug url for more details. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 542 (write-behind crashes) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=542
* performance/write-behind: refactor __wb_collapse_write_bufsRaghavendra G2010-02-221-36/+38
| | | | | | | | | | - refactor done to remove duplicated code. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 542 (write-behind crashes) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=542
* common-utils: gf_log2 handles 0 inputAnand Avati2010-02-221-1/+1
| | | | | | | | | Signed-off-by: Anand V. Avati <avati@amp.gluster.com> Signed-off-by: Anand V. Avati <avati@blackhole.gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 583 (filesystem access hangs while deleting large files) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=583
* write-behind: fix 'option enable-trickling-writes'Anand Avati2010-01-261-8/+4
| | | | | | | | | | | fix logic in __wb_mark_winds to properly initialize incomplete_writes and conditionally check based on enable_trickling_writes flag Signed-off-by: Anand V. Avati <avati@amp.gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 547 (write-behind enable-trickling-writes working?) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=547
* Added null checks in "fini"Harshavardhana Ranganath2009-12-182-0/+6
| | | | | | | | Signed-off-by: Harshavardhana <harsha@gluster.com> Signed-off-by: Vijay Bellur <vijay@dev.gluster.com> BUG: 364 (Segfault in io-cache) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=364
* Revert "performance/write-behind: Changed default value of option ↵Anand V. Avati2009-12-061-1/+1
| | | | | | | | | disable-till to 0" This reverts commit e6c35f74ff4a204cf016ebf201a2630b331f9434. This commit is found to trigger spurious self heal by causing flush and writev to get reordered. Reverting as a workaround.
* performance/stat-prefetch: remove the usage of THIS.Raghavendra G2009-12-062-1/+3
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: fix memory leak.Raghavendra G2009-12-061-1/+4
| | | | | | | | | | | - rbt_hash_remove only removes entry from the hash table. It does not free the data. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 438 (Memory leak) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=438
* performance/stat-prefetch: don't check whether previous lookups have been ↵Raghavendra G2009-12-051-9/+0
| | | | | | | | | | | | | | | | | | | | failed in sp_process_inode_ctx. We need this patch for following reasons: 1. The inode used in the successful lookup call may not be the one passed to fops operating on path. This may happen, if revalidate of inode is failed, but the lookup itself succeeds (in case of reavalidate failures, fuse sends a fresh lookup using new inode) and since inode_link, if there is an inode already associated with a path retains it. 2. Only fops that "wait" in stat-prefetch for a lookup which is already in progress, need to check for the result of the lookup they are waiting on, before resuming and sp_process_inode_ctx is _not_ used in such contexts. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 428 (looks like stat cache not upto date..) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=428
* performance/stat-prefetch: add "caller" arguement to sp_process_inode_ctx.Raghavendra G2009-12-041-21/+37
| | | | | | | | | | | - since sp_process_inode_ctx is called by many fops, if the context is not set "caller" argument can be logged in log msgs to identify caller. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: implement sp_check_and_create_inode_ctx.Raghavendra G2009-12-041-161/+122
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: refactor sp_lookup_cbk to use sp_update_inode_ctx.Raghavendra G2009-12-041-29/+10
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: dont check for inode context in fops like create, ↵Raghavendra G2009-12-041-259/+262
| | | | | | | | | | | | | mkdir. - context will not be set as a fresh inode is passed in each of these calls. instead create a new context. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: don't access stat if lookup has been failed.Raghavendra G2009-12-031-1/+1
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 429 (crash in lookup_cbk when lookup returns with error) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=429
* performance/write-behind: Changed default value of option disable-till to 0Vijay Bellur2009-12-021-1/+1
| | | | | | | | Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 425 (Change disable-till default to 0) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=425
* performance/stat-prefetch: get inode context from local->loc.inode instead ↵Raghavendra G2009-12-021-4/+9
| | | | | | | | | | | | of inode passed as arguement in sp_lookup_cbk. - @inode in sp_lookup_cbk may be NULL in case of failure of lookup. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 426 (stat on mount point hangs.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=426
* performance/stat-prefetch: fix access to inode table.Raghavendra G2009-12-021-13/+23
| | | | | | | | | | | - Access the inode table from one of fd->inode->table, inode->table or parent->table instead of from root xlators itable. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 369 (Samba does not work with booster.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=369
* fixes to compile on MacOSX (no fuse client)Amar Tumballi2009-12-012-12/+13
| | | | | | | | | | | | | | | These changes are required to make GlusterFS compile on MacOSX (10.5). Currently glusterfs server component alone will work over Mac, and it has to be built with following options to ./configure. "bash$ ./configure --disable-fuse-client --disable-fusermount " Signed-off-by: Amar Tumballi <amar@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 361 (GlusterFS 3.0 should work on Mac OS/X) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=361
* performance/io-cache: Hold locks on inode before creating rbt tree.Vijay Bellur2009-12-011-20/+29
| | | | | | | | | | Lock is held before creating a RBT in inode. Also, did some re-factoring. Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 335 (Io-cache optimization) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=335
* stat-prefetch: fix unsafe reference to @local in sp_lookup_cbkAnand Avati2009-12-011-1/+5
| | | | | | | | | | | | | stat-prefetch uses the main syscall frame for performing lookup on demand. this causes a potential reference to a freed local in sp_lookup_cbk after resuming the main syscall frame (which could have unwound and destroyed) Signed-off-by: Anand V. Avati <avati@blackhole.gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 242 (If any of the writes fail, write-behind should not wait till the fd is closed for reporting errors) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=242
* performance/stat-prefetch: make lookup to wait for the completion of another ↵Raghavendra G2009-12-011-14/+106
| | | | | | | | | | | | | | | | lookup on same path if one is in progress. - If current lookup (2) does not wait for completion of the lookup (1) which is in progress, there can be a race condition where (2) completes ahead of (1) and resuming all the waiting operations in the queue. When (1) returns, the original operation (eg., stat, chmod etc) might've already unwound and hence the frame would've been destroyed. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: don't use same variable for getting inode context ↵Raghavendra G2009-12-011-5/+5
| | | | | | | | | | | | | of an inode and its parent. - this bug results in updation of flags in inode context of the parent instead of that of inode. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/stat-prefetch: fix errors in handling memory allocation failures ↵Raghavendra G2009-12-011-3/+3
| | | | | | | | | | | | | | in sp_inode_ctx_process. - the variables that are used for making decisions are assigned before memory allocations and these variables were not reset to correct values in case of memory allocation failures. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* performance/quick-read: change the maximum value of max-file-size to 1000KB ↵Raghavendra G2009-11-281-4/+4
| | | | | | | | | | | | | | | from 1MB. - protocol restricts the maximum buffer size that can be received over network to 1MB. Since the buffer holding reply of lookup should also contain the lookup header and other key/value pairs of xattr_reply dict, the maximum file size that can be fetched through dictionary cannot be equal to 1MB. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 419 (Issues with writing files of size >= 1MB) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=419
* performance/stat-prefetch: Changed rbthash to make use of 1 bucketVijay Bellur2009-11-262-16/+66
| | | | | | | | | | rbthash makes use of 1 bucket and a common mem-pool is being used for all rbt entries. Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 221 (stat prefetch implementation) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221
* Changed rbthash_table_init() to take a mem-pool argument.Vijay Bellur2009-11-264-20/+66
| | | | | | | | | | | | | | | | | | | Changes in libglusterfs/rbthash: rbthash_table_init() now takes a mem-pool argument. The mem-pool argument would be mutually exclusive to expected_entries. If expected_entries is provided, mem-pool would be ignored and vice-versa. Changes in io-cache: 1) Moved rbthash creation to readv. 2) rbthash makes use of 1 rbt instead of 4096 3) A global mem-pool is being used in place of a mem-pool per rbt. Signed-off-by: Vijay Bellur <vijay@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 335 (Io-cache optimization) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=335
* performance/stat-prefetch: Add refcounting to cache objectsRaghavendra G2009-11-241-6/+88
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 405 (Segmentation fault in stat-prefetch.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=405
* performance/stat-prefetch: remove lock member in sp_fd_ctx_t.Raghavendra G2009-11-242-36/+22
| | | | | | | | | | | - since the lifetime of fd context can be no longer than the fd, we can use lock in fd. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 405 (Segmentation fault in stat-prefetch.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=405
* performance/stat-prefetch: don't free the cache in readdir if the offset is ↵Raghavendra G2009-11-241-5/+2
| | | | | | | | | | | | | | | | | | | | | | not the expected one. - cache creation is expensive operation. Also, cache will be freed in releasedir. Hence, just remove all entries from cache without freeing the cache. However this is not entirely true, since sp_cache_remove_entry frees the old table and reinitializes a new table if all entries are being removed. When rbtree based hash table provides an interface to remove all the entries, sp_cache_remove_entry should be modified not to destroy the table. - this patch also fixes a race condition wherein the cache being used in lookup getting freed in readdir if the offset is not equal to expected offset. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 405 (Segmentation fault in stat-prefetch.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=405
* performance/quick-read: use STACK_UNWIND_STRICT for unwinding.Raghavendra G2009-11-241-40/+49
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
* performance/write-behind: use STACK_UNWIND_STRICT for unwinding.Raghavendra G2009-11-241-2/+2
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
* performance/read-ahead: use STACK_UNWIND_STRICT for unwinding.Raghavendra G2009-11-242-18/+20
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
* performance/io-threads: use STACK_UNWIND_STRICT for unwinding.Raghavendra G2009-11-241-90/+102
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
* performance/io-cache: use STACK_UNWIND_STRICT for unwinding.Raghavendra G2009-11-242-20/+26
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
* performance/stat-prefetch: use STACK_UNWIND_STRICT for unwinding.Raghavendra G2009-11-242-93/+105
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 269 (Add a specialized STACK_UNWIND macro for each FOP) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=269
* performance/io-cache: don't use stat got in read_cbk if it is zero-filled.Raghavendra G2009-11-202-16/+29
| | | | | | | | | | | | - translators like io-cache, read-ahead return a zero-filled stbuf in readv_cbk and usage of zero filled stat for cache validation is not correct. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 376 (server-side io-cache is preventing client-side io-cache from working) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=376
* performance/stat-prefetch: don't cache dentries corresponding to directory.Raghavendra G2009-11-191-0/+4
| | | | | | | | | | | | | - distribute picks up the inode number from hashed subvolume. But readdir is done from different subvolumes based on the offset. Hence there is a possibility of inode number and device number for directories being inconsistent between two stat calls. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 371 (rm -rf fails on stat-prefetch.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=371
* performance/stat-prefetch: don't use pre-allocated dirent structure in ↵Raghavendra G2009-11-191-8/+23
| | | | | | | | | | | | | | sp_get_cache_entry. - since we don't know the size of dentry before calling sp_get_cache_entry, we must dynamically allocate the dentry in sp_cache_get_entry and copy the contents from cache. Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 371 (rm -rf fails on stat-prefetch.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=371
* performance/quick-read: set return value to zero after successful init.Raghavendra G2009-11-131-0/+2
| | | | | | | | Signed-off-by: Raghavendra G <raghavendra@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 374 (quick read fails to initialize if no options are given) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=374
* performance/stat-prefetch: In lookup_cbk do not set error no to EINVAL, when ↵vinayak hegde2009-11-051-3/+0
| | | | | | | | | | inode_ctx_get returns -1. Signed-off-by: vinayak <vinayak@laptop.(none)> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 360 (All fop fails when stat-prefetch is loaded on afr.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=360
* performance/stat-prefetch: In sp_readdir_cbk op_ret value is not handled.vinayak hegde2009-11-051-0/+4
| | | | | | | | Signed-off-by: Vinayak Hegde <vinayak@gluster.com> Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 360 (All fop fails when stat-prefetch is loaded on afr.) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=360
* performance/io-cache: change the data-types to be big enough to hold size of ↵Raghavendra G2009-10-302-5/+5
| | | | | | | | | data flushed. Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 335 (Io-cache optimization) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=335
* performance/io-cache: change the hash function used for rbtree based hash table.Raghavendra G2009-10-292-12/+16
| | | | | | | | | | | | | | | - the earlier hash function does not distribute pages uniformly for offsets that fit into 32 bits. The reason is that the hash function just xors the contents of the key 4 bytes at a time with the current value of hash. Hence for keys that fit into 32 bits, the hash will be the key itself. Since we are using the rounded_offset (which is a multiple of 128KB) as the key, the key will be exactly divisible by the number of buckets configured (4096) resolving all the pages into the first bucket. Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 335 (Io-cache optimization) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=335
* libglusterfs/rbtree: change rbthash_init_table to take no of expected ↵Raghavendra G2009-10-282-6/+14
| | | | | | | | | | | | | | entries in the hash table as argument. - the expected number of entries is used to create the memory pool of the hash table. Having constant macro for this purpose is not suitable since different users of rbtree based hash table store different number of entries in the table. Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 335 (Io-cache optimization) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=335
* performance/io-cache: fix memory leak.Raghavendra G2009-10-281-0/+1
| | | | | | | | | - page table of cache was not being destroyed in ioc_forget. Signed-off-by: Anand V. Avati <avati@dev.gluster.com> BUG: 335 (Io-cache optimization) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=335
generated by cgit (git 2.34.1) at 2024-05-14 06:19:35 +0000