[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
[aspectj-users] Re: Redistribution of aspectj 1.6.7.a as, 1.6.8?
|
Hi Andy,
Here are some figures from our comparison of 1.6.6 with the latest dev
build (1.6.7a+). We do 3 tests with jboss application server with all
classes in org.jboss instrumented but none of the applications themselves.
The first test is performed without our front bytecode cache. The second
with the cache enabled which is empty for this run but is persisted on
shutdown for the next run. The third with the cache enabled and the
previous instrumentation images loaded - effectively switching to
compile/build instrumentation mode.
I measure the memory after the server has started up completed and then
again after a few GC's have been initiated.
1.6.6
Without JXInsight Cache
32 secs 208 MB 168 MB
With JXInsight Cache #1
34 secs 188 MB 171 MB
With JXInsight Cache #2
21 secs 115 MB 108 MB
1.6.7a (latest dev build)
Without JXInsight Cache
33 secs 170 MB 139 MB
With JXInsight Cache #1
34 secs 171 MB 147 MB
With JXInsight Cache #2
21 secs 117 MB 109 MB
I do not think our tests currently take advantage of the matching
optimizations because of that exclude clause that uncovered the issue
with 1.6.7. I will try to get a build prepared without this to see the
performance benefits.
But it does look like the memory footprint has been greatly reduced.
Kind regards,
William
On 1/7/10 7:23 PM, William Louth (JINSPIRED.COM) wrote:
Date: Wed, 6 Jan 2010 10:33:35 -0800
From: Andy Clement<andrew.clement@xxxxxxxxx>
Subject: Re: [aspectj-users] Redistribution of aspectj 1.6.7.a as
1.6.8?
Interesting. We've had qualifiers previously, but not since the early
1.5 days if i recall correctly so I guess that hasn't come up before.
I haven't actually put 1.6.7.a in a repo yet as I'm waiting on a few
more people to report back - it seems although I kept the code stable
for a few weeks before release and asked users to try it, they didn't
get around to it (I'm not blaming anyone, it is just unfortunate) and
now problems are being reported on the final release. There is the
one fix I've made so far to create 1.6.7.a but Simone has perhaps seen
another issue and a user on the spring forums is describing an old
problem which seems to occur more frequently on 1.6.7 (I'm working on
it under https://bugs.eclipse.org/bugs/show_bug.cgi?id=298908 ).
Depending on what happens with these, I may create a 1.6.8 release
anyway, rather than getting into b/c/d suffixes. I'll decide what to
do in the next few days.
Can I ask anyone who is able to, please try 1.6.7.a and raise any
issues you see.
Andy
I assume Andy you want us to try out the latest development build
which includes the fix isAssignableFrom which occurred previously when
WeakReferences were used liberally and heap memory was low (which
might explain why it is always hard to create a test case for).
I would also like to see it named 1.6.8 though I understand why you
would not like to jump so quickly after a release pushed out a week
ago (except for me).
Kind regards,
William
