path: root/doc/booster.txt
diff options
authorVikas Gorur <>2009-02-18 17:36:07 +0530
committerVikas Gorur <>2009-02-18 17:36:07 +0530
commit77adf4cd648dce41f89469dd185deec6b6b53a0b (patch)
tree02e155a5753b398ee572b45793f889b538efab6b /doc/booster.txt
parentf3b2e6580e5663292ee113c741343c8a43ee133f (diff)
Added all files
Diffstat (limited to 'doc/booster.txt')
1 files changed, 54 insertions, 0 deletions
diff --git a/doc/booster.txt b/doc/booster.txt
new file mode 100644
index 0000000..684ac89
--- /dev/null
+++ b/doc/booster.txt
@@ -0,0 +1,54 @@
+* booster is a LD_PRELOADable library which boosts read/write performance by bypassing fuse for
+ read() and write() calls.
+* fetch volfile from glusterfs.
+* identify whether multiple files are from the same mount point. If so, use only one context.
+* for a getxattr, along with other attributes, fuse returns following attributes.
+ * contents of client volume-file.
+ * mount point.
+* LD_PRELOADed maintains an hash table storing mount-points and libglusterfsclient handles
+ so that handles are reused for files from same mount point.
+* it also maintains a fdtable. fdtable maps the fd (integer) returned to application to fd (pointer to fd struct)
+ used by libglusterfsclient. application is returned the same fd as the one returned from libc apis.
+* During fork, these tables are overwritten to enable creation of fresh glusterfs context in child.
+* application willing to use booster LD_PRELOADs which is a wrapper library implementing
+ open, read and write.
+* application should specify the path to logfile through the environment variable GLFS_BOOSTER_LOGFILE. If
+ not specified, logging is done to /dev/stderr.
+* open call does,
+ * real_open on the file.
+ * fgetxattr(fd).
+ * store the volume-file content got in the dictionary to a temparory file.
+ * look in the hashtable for the mount-point, if already present get the libglusterfsclient handle from the
+ hashtable. Otherwise get a new handle from libglusterfsclient (be careful about mount point not present in
+ the hashtable and multiple glusterfs_inits running simultaneously for the same mount-point there by using
+ multiple handles for the same mount point).
+ * real_close (fd).
+ * delete temporary volume-volfile.
+ * glusterfs_open (handle, path, mode).
+ * store the fd returned by glusterfs_open in the fdtable at the same index as the fd returned by real_open.
+ * return the index as fd.
+* read/write calls do,
+ * get the libglusterfsclient fd from fdtable.
+ * if found use glusterfs_read/glusterfs_write, else use real_read/real_write.
+* close call does,
+ * remove the fd from the fdtable.
+* other calls use real_calls.