bk commit into 5.1 tree (jonas:1.2080)

[jonas]

Below is the list of changes that have just been committed into a local
5.1 repository of jonas. When jonas does a push these changes will
be propagated to the main repository and, within 24 hours after the
push, to the public repository.
For information on how to access the public repository
see http://dev.mysql.com/doc/mysql/en/installing-source-tree.html

ChangeSet@stripped, 2006-11-01 14:22:25+01:00, jonas@stripped +2 -0
  drop6p0-alpha

  cluster_change_hist.txt@stripped, 2006-11-01 14:22:23+01:00, jonas@stripped +34 -0
    update change log

  wan.txt@stripped, 2006-11-01 14:21:54+01:00, jonas@stripped +20 -0
    BitKeeper file /home/jonas/src/mysql-5.1-wl2325-5.0-drop6/wan.txt

  wan.txt@stripped, 2006-11-01 14:21:54+01:00, jonas@stripped +0 -0

# This is a BitKeeper patch.  What follows are the unified diffs for the
# set of deltas contained in the patch.  The rest of the patch, the part
# that BitKeeper cares about, is below these diffs.
# User:	jonas
# Host:	perch.ndb.mysql.com
# Root:	/home/jonas/src/mysql-5.1-wl2325-5.0-drop6

--- 1.7/cluster_change_hist.txt	2006-11-01 14:22:28 +01:00
+++ 1.8/cluster_change_hist.txt	2006-11-01 14:22:28 +01:00
@@ -9,6 +9,40 @@
 (note that this does not apply to the not yet released Alcatel features for Partitioning and Replication)
 (* == Alcatel has hit this bug)
 
+
+
+-------------------------------
+
+Alcatel Drop 6p0-alpha (Fri: 1 Nov 2006)
+
+As drop5p16b with additional features:
+
+    * READ ALL - If application reads all columns from table
+                 (doing getValue in column order) only 32 bit attrinfo
+                 is sent from API to DN, instead of #columns
+    * READ OPT. - If read, a bitmap of columns to read is sent
+                  from API to DN instead of 32-bit per column
+    * RESULT SET OPT.
+                - In datasent from DN to API, a bit map of which columns
+                  has been read is sent instead of 32-bit per column
+    * VARSIZE RESULT SET
+                - When sending varsize data from DN to API only real length
+                  is sent instead of max length.
+
+                  Note this requires that varsize columns are packed correctly
+                  i.e first byte = length if varbinary
+                      first two bytes = length if longvarbinary (little endian)
+
+ Note: All above mentioned optimizations are currently only applied to
+       primary key operations.
+
+    * 255 API Nodes
+    * Possibility to configure nodes as Wan node.
+      Add a "Wan: yes" in config.ini and buffers for TCP connection will
+      be set higher than for non-wan nodes.
+
+      See separate document on how to configure your linux-os to support this
+
 -------------------------------
 
 Alcatel Drop 5p16b (Fri: 23 Oct 2006)
--- New file ---
+++ wan.txt	06/11/01 14:21:54
    It is quite likely that you need to change the defaults on the OS
    to get optimal performance on WAN links. The socket options cannot
    set the value higher than the configuration variable rmem_max and
    wmem_max.
    E.g. in Fedora Core 5 these are changed by writing the following
    lines in the file /etc/sysctl.conf:
    net.core.rmem_max=4194304
    net.core.wmem_max=4194304
    This sets the maximum value to 4 MBytes.
    It is also required to set tcp_mem, tcp_rmem and tcp_wmem to proper
    values in those cases.  E.g again in /etc/sysctl.conf on Fedora Core 5
    net.ipv4.tcp_mem= 16384 174670 4194304
    net.ipv4.tcp_rmem= 16384 65536 4194304
    net.ipv4.tcp_wmem= 16384 65536 4194304
    The first parameter says the minimum size of the socket buffer in
    situations with high memory pressure, the second is the default
    size of the buffer and the last parameter is the maximum size it
    can be set to. This cannot be set higher than the values above so
    these must also be set.