Erlang Halfword 测试

[quote]
OTP-8941 == erts ==
From this release, the previously experimental halfword
emulator is now official. It can be enabled by giving the
--enable-halfword-emulator option to the configure script.

The halfword emulator is a 64-bit application, but uses
halfwords (32-bit words) for all data in Erlang processes,
therefore using less memory and being faster than the
standard 64-bit emulator. The total size of all BEAM code and
all process data for all processes is limited to 4Gb, but ETS
tables and off-heap binaries are only limited by the amount
of available memory.
[/quote]

看上面的Erlang的halfword描述，貌似halfword特性很给力啊。对于Processes的寻址是32位，以为着更少的内存，更快的访问，受到4Gb的限制。可是对于tables（ets/mnesia）和off-heap binaries是64位的寻址。

我们来测试一下省了多少内存：

1、 ./configure; make ; sudo make install (64bit version)

erl +A 500 +K true

1> mnesia:start(), 
1> rd(test, {k,v}), 
1> mnesia:create_table(test, [{ram_copies, [node()]}, {attributes, record_info(fields, test)}  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
11879 witeman   20   0 4269m  77m 2408 S    0  2.0   0:14.90 beam.smp           
]),
1> lists:foreach(fun(X) -> mnesia:d  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
11879 witeman   20   0 4269m  77m 2408 S    0  2.0   0:14.90 beam.smp           
irty_write(#test{k = X, v = X}) end, lists:seq(1000001,2000000)),
1> mnesia:table_info(test, memory).
8144942
2> mnesia:info().
test           : with 1000000  records occupying 8144942  words of mem
schema         : with 2        records occupying 522      words of mem
3> memory().      
[{total,116395728},
 {processes,26512432},
 {processes_used,26493672},
 {system,89883296},
 {atom,606233},
 {atom_used,578661},
 {binary,739408},
 {code,5069980},
 {ets,81482024}]

top:
PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
 9625 witeman   20   0  322m 113m 2424 S    0  2.9   0:29.57 beam.smp    

2、./configure --enable-halfword-emulator; make ; sudo make install (64bit halfword version)

1> mnesia:start(),
1> rd(test, {k,v}),
1> mnesia:create_table(test, [{ram_copies, [node()]}, {attributes, record_info(fields, test)}]),
1> lists:foreach(fun(X) -> mnesia:dirty_write(#test{k = X, v = X}) end, lists:seq(1000001,2000000)),mnesia:table_info(test, memory).
11289878
2> mnesia:info().8144942
test           : with 1000000  records occupying 11289878 words of mem
schema         : with 2        records occupying 822      words of mem
3> memory().
[{total,73691680},
 {processes,916040},
 {processes_used,903440},
 {system,72775640},
 {atom,602993},
 {atom_used,573864},
 {binary,70896},
 {code,4821930},
 {ets,65397888},
 {low,5418320}]
4> erlang:system_info(wordsize).  
4

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
11879 witeman   20   0 4269m  77m 2408 S    0  2.0   0:14.90 beam.smp           

mnesia对比:
64bit version: 8144942 * 8 bytes = 65159536 bytes
64bit halfword version: 11289878 * 4 = 45159512 bytes

rate = 45159512 / 65159536 = 69.31%，也就是说halfword对于mnesia里的integer能减少30.69%的内存。

total memory对比：
64bit version: 116395728 bytes
64bit halfword version: 73691680 bytes

rate = 73691680 / 116395728 = 63.31%，也就是说halfword对于整个Erlang VM来说能减少36.69%的内存使用


整个VM内存使用在OS层面的对比：
64bit version: 112 m
64bit halfword version: 77 m

rate = 77 / 112 = 68.75%，也就是说halfword对于VM在OS层面上来说能减少31.25%的内存使用