[webstack-discuss] Enhancing Ruby with DTrace probes [PSARC/2008/174 Self Review]

[Prashant Srinivasan]

Richard Lowe wrote:
> Jyri Virkki <jyri at sac.sfbay.sun.com> writes:
>
>   
>> I am sponsoring this case for Sivakumar and closing it approved automatic.
>> The only high[low]light is the feature disparity between x86 and SPARC.
>> If anyone would like to discuss it as a fast track let me know.
>>
>>     
>
> It'd be nice if the project team could go into why the disparity
> between x86 and SPARC exists.  You mention it, both here and in the
> materials but I don't see any description of why these probes can't also be
> delivered for SPARC.
>
> Is this an actual decision to not deliver on sparc? does something
> prevent delivery on sparc that you can't wait to resolve?
>   


Hi Rich,
 The project aims for DTrace support on SFWNV-Ruby-x86 and 
SFWNV-Ruby-SPARC.  The DTrace probes break the SFWNV Ruby build on 
SPARC.  The x64 build works as expected.

 We investigated the problem, the source code looks fine, it appears to 
be a bug with the compiler(6659110).  We're actually able to get the 
SPARC build to work, if we remove all compiler flags _except_ "-g".  
(ie., CFLAGS=-g is the only way to build with SS11).

 We investigated that option as well - we ran microbenchmarks (both 
home-made and from the Debian Ruby Alioth tests), and saw an average 
performance degradation of about 30% with this workaround(I can send out 
the individual performance test details if needed).

 So, the question was whether we provide DTrace functionality on SPARC 
at the cost of having Ruby/SPARC run a third slower . . . and we 
decided, for the time being, to wait for the bug 6659110 to be resolved 
before integrating DTrace probes in SPARC.

 We're definitely committed to delivering DTrace support for SFWNV Ruby 
on SPARC.

 We've gotten requests for DTrace support in Ruby from Sun and non-Sun 
sources . . . and at least one other OS has Ruby/DTrace support bundled 
in - hence we moved to get the probes into x64 without waiting on the 
compiler bug to be fixed.

 Hope that sets some context to this case.
 -ps





> -- Rich
>
>   






-- 
Prashant Srinivasan
F/OSS Enthusiast
Sun Microsystems, Inc.
http://blogs.sun.com/prashant
GnuPG key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x82FBDE5A