Review/comment needed for the new public java.util.Base64 class (original) (raw)

Xueming Shen xueming.shen at oracle.com
Sat Oct 20 00:40:04 UTC 2012


Understood. I do have the code:-) but I'm hesitated to go SharedSecrets simply for performance gain of a utility method. This definitely can be addressed if it turns out to be a real issue standing in critical path.

-Sherman


sherman at sherman-linux:~/Workspace/jdk8/test/java/util/Base64$ java
PermBase64 200000 1000 j.u.Base64.encode(ba) : 528745 j.u.Base64.encodeString(ba): 739703 j.u.Base64.encode(bb) : 486216 j.u.Base64.encode(bb, bb) : 538544 j.u.Base64.encode(bb, bb)-D: 850947 migBase64.encode(ba) : 659474 vs

sherman at sherman-linux:~/Workspace/jdk8/test/java/util/Base64$ java
PermBase64 200000 1000 j.u.Base64.encode(ba) : 519391 j.u.Base64.encodeString(ba): 964854 j.u.Base64.encode(bb) : 490138 j.u.Base64.encode(bb, bb) : 539027 j.u.Base64.encode(bb, bb)-D: 786438 migBase64.encode(ba) : 660572

-Sherman

On 10/19/2012 04:59 PM, Mike Duigou wrote:

For me the greater concern, which is hard to measure, is the GC pressure added by the discarded byte array.

Mike On Oct 19 2012, at 17:03 , Xueming Shen wrote:

I see a 20% performance gain on server vm if switch to pure char[] based encoding and then use the sharedSecrets to avoid the copy. The dis-advantage is (1) have to use the sharedSecrets and (2) can't share the same between the encode(byte[]) and encode(String).

Anyway it appears to be an alternative for performance improvement. -Sherman

On 10/18/2012 01:07 PM, Mike Duigou wrote: I wonder if there would be advantage in using a SharedSecrets mechanism to allow construction of a String directly from a char array. The intermediate byte array seems wasteful especially for what is likely to be a heavily used path. Mike On Oct 17 2012, at 19:10 , Xueming Shen wrote:

Hi

Webrev has been updated with following changes (1) added a pair of en/decode(ByteBuffer src, ByteBuffer dst) methods (2) some minor spec clarification regarding the "end of decoding" (3) performance tuning. webrev: http://cr.openjdk.java.net/~sherman/4235519/webrev some performance scores: http://cr.openjdk.java.net/~sherman/4235519/score3 -Sherman



More information about the core-libs-dev mailing list