Import from TextPattern

diff --git a/news/102.mdwn b/news/102.mdwn
new file mode 100644 (file)
index 0000000..fee135c
--- /dev/null
+++ b/news/102.mdwn
@@ -0,0 +1,31 @@
+[[!meta title="Java no longer broken on scripts.mit.edu"]]
+Yesterday it was brought to our attention that our resource limits
+were preventing Java from working, because the JVM attempted to
+allocate more memory than our limits.
+
+Because of problems in the past with Java instances using 2 GB or
+more memory causing out-of-memory conditions, we recently
+reimplemented a 1 GB per-process [resource limit](/faq/34/).
+However, the Java startup code’s attempt at allocating a
+“reasonable fraction” of the total physical memory on the server
+makes it attempt to grab about 1.05 GB. (This is a
+[known issue](http://bugs.sun.com/view_bug.do?bug_id=6374896) in
+Java.) This amount might be reasonable for a server running nothing
+but Java, but is entirely too much for a shared host like
+scripts.mit.edu. So, we’re setting the JAVA\_TOOL\_OPTIONS=-Xmx128M
+environment variable to limit the maximum Java heap size, and Java
+should be working again.
+
+If this isn’t enough for your application, you can create a
+**.hotspotrc** file in the same directory as your script or Java
+code, containing the option **MaxHeapSize=256M** (or possibly 512M;
+note that we’ve tested 768M and found it to be unstable), or you
+can pass **-Xmx256M** on your Java command line.
+
+It looks like we don’t have very many Java users at all because
+this problem went unreported, but if you have any questions, as
+always, please let us know at
+[scripts@mit.edu](mailto:scripts@mit.edu).
+
+
+