Why is default Young Space so small?

[Shane Cox]

We recently moved one of our apps onto a new box running i7 processors
(Nehalem).  After doing so, we noticed that the default Young Space
allocation decreased substantially even though the new box has more
processors and RAM ... and the JVM allocated more Parallel GC threads.

Old machine:
# mpstat
CPU minf mjf xcal  intr ithr  csw icsw migr smtx  srw syscl  usr sys  wt idl
  0    3   0   49  2355 1251  274    3   15    2    0   345    1   1   0  99
  1    3   0   48   188    0  284    4   17    2    0   359    1   0   0  99
  2    7   0   58   238  104  214    3   12    4    0   294    0   0   0  99
  3    2   0   29   127    3  144    2    8    1    0   257    0   0   0  99

# prtconf
System Configuration:  Sun Microsystems  i86pc
Memory size: 8192 Megabytes

Start of GC log shows 59MB Young Space:
3.242: [GC 3.242: [ParNew: 52480K->6528K(*59008K*), 0.0221415 secs]
52480K->7228K(761472K), 0.0222595 secs]



New machine:
# mpstat
CPU minf mjf xcal  intr ithr  csw icsw migr smtx  srw syscl  usr sys  wt idl
  0  229   0  686   448  142  221    0    3   10    0   662    0   1   0  99
  1  181   0  593   114    0  188    0    1   10    0   574    0   0   0  99
  2   16   0   88   148    0  230    0    1    7    0   478    1   0   0  99
  3   16   0  199   161   21  215    0    0    9    0   532    0   0   0  99
  4    5   0   50   150    0  181    0    1    5    0   378    0   0   0  99
  5   21   0  125   147    2  184    0    0   10    0   523    0   0   0  99
  6    2   0  185   145    4  157    0    1    3    0   317    0   0   0  99
  7   16   0  114   152    7  128    0    0    5    0   367    0   0   0 100
  8    1   0   73   456  307   45    0    1    2    0    50    0   0   0 100
  9   15   0   86   148    0   12    0    0    1    0    89    0   0   0 100
 10    0   0   15   158    6   26    0    1    1    0    28    0   0   0 100
 11    2   0   31   151    0    9    0    0    1    0    36    0   0   0 100
 12    0   0    7   153    0   12    0    0    1    0    19    0   0   0 100
 13    1   0    8   153    0    4    0    0    0    0    55    0   0   0 100
 14    0   0   15   152    0    5    0    0    1    0    14    0   0   0 100
 15    1   0    6   153    0    7    0    0    1    0    21    0   0   0 100

# prtconf
System Configuration:  Sun Microsystems  i86pc
Memory size: 24566 Megabytes

Start of GC log shows approx 15MB Young Space:
0.543: [GC 0.543: [ParNew: 13184K->1483K(*14784K*), 0.0029503 secs]
13184K->1483K(766400K), 0.0030164 secs] [Times: user=0.02 sys=0.00,
real=0.00 secs]


Java version is the same on both machines:
java version "1.6.0_14"
Java(TM) SE Runtime Environment (build 1.6.0_14-b08)
Java HotSpot(TM) Server VM (build 14.0-b16, mixed mode)



I did a thread dump on the new machine to determine how many Parallel GC
Threads (Gang Workers) are running.  Looks like 13:
"Gang worker#0 (Parallel GC Threads)" prio=3 tid=0x08075400 nid=0x3 runnable

"Gang worker#1 (Parallel GC Threads)" prio=3 tid=0x08076400 nid=0x4 runnable

"Gang worker#2 (Parallel GC Threads)" prio=3 tid=0x08077800 nid=0x5 runnable

"Gang worker#3 (Parallel GC Threads)" prio=3 tid=0x08078c00 nid=0x6 runnable

"Gang worker#4 (Parallel GC Threads)" prio=3 tid=0x0807a000 nid=0x7 runnable

"Gang worker#5 (Parallel GC Threads)" prio=3 tid=0x0807b800 nid=0x8 runnable

"Gang worker#6 (Parallel GC Threads)" prio=3 tid=0x0807cc00 nid=0x9 runnable

"Gang worker#7 (Parallel GC Threads)" prio=3 tid=0x0807e000 nid=0xa runnable

"Gang worker#8 (Parallel GC Threads)" prio=3 tid=0x0807f400 nid=0xb runnable

"Gang worker#9 (Parallel GC Threads)" prio=3 tid=0x08080800 nid=0xc runnable

"Gang worker#10 (Parallel GC Threads)" prio=3 tid=0x08081c00 nid=0xd
runnable
"Gang worker#11 (Parallel GC Threads)" prio=3 tid=0x08083000 nid=0xe
runnable
"Gang worker#12 (Parallel GC Threads)" prio=3 tid=0x08084400 nid=0xf
runnable



We use the CMS collector.  We don't specify a NewSize.  We like to run with
defaults if possible.


Any ideas why the JVM decreased the Young Space allocation by 3/4th even
though the number of processors increased by 4x?

Thanks
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-gc-dev/attachments/20091016/a08128ba/attachment.htm>