JEP 291: Deprecate the Concurrent Mark Sweep (CMS) Garbage Collector

kirk at kodewerk.com kirk at kodewerk.com
Fri Jul 1 22:33:57 UTC 2016


Hi Jon,

Thank you for the link. I have some technical arguments based on my experiences moving clients to G1 that suggest that the G1, in it’s current form, isn’t always a great replacement for CMS. I’ll add them to the JEP.

As an FYI, I’ve written a little FX app that visualizes region layout in the G1 using the PrintRegionInformation flag up on GitHub. I’ve been visualizing data from live production environment and it’s very interesting to see how humungous allocations sit in memory. As you most likely know, humungous allocations are but one of the things that can give you some grief.

Kind regards,
Kirk Pepperdine

> On Jul 1, 2016, at 10:26 PM, Jon Masamitsu <jon.masamitsu at oracle.com> wrote:
> 
> Kirk,
> 
> On 7/1/2016 12:19 PM, kirk at kodewerk.com wrote:
>> Hi Jon,
>> 
>> I’m beyond convinced that you understand how CMS functions which left me wondering what is really going on here. I’m not really concerned about what Oracle decides to put into it’s builds. However, there are other parties vested in OpenJDK which implies that pulling CMS from OpenJDK needs to be a completely different conversation. Who do we need to speak to?
> 
> Add your comments to the JEP.
> 
> https://bugs.openjdk.java.net/browse/JDK-8142518
> 
> The JEP lays out why Oracle thinks dropping CMS support is a good
> idea.   If you can suggest an alternative that gets us some of what
> Oracle wants and gets the community what it wants, that would
> have more weight than just disagreeing with the  reasoning there.
> 
> Much of what I think Oracle wants is to spend the engineering
> dollars (engineers, hardware, answering the customer calls, etc.)
> on G1 instead of CMS.  We don't spend much development time
> on CMS now but bug fixing,  testing and support goes on and on.
> I think that has to be addressed.
> 
> This is an invitation to anyone to address this issue with an
> alternative plan.
> 
> I'll get someone to respond to your comments.
> 
> Jon
> 
>> 
>> Kind regards,
>> Kirk Pepperdine
>> 
>>> On Jul 1, 2016, at 8:38 PM, Jon Masamitsu <jon.masamitsu at oracle.com> wrote:
>>> 
>>> Kirk,
>>> 
>>> Thanks for your email.   Please understand that I'm not the one
>>> that needs to be convinced that CMS could never be turned on.
>>> And maybe CMS being turned on isn't even the first or last
>>> question.   I'm guessing including the CMS bits in an Oracle
>>> build is a decision that would have to be made at a fairly
>>> high management level.
>>> 
>>> Jon
>>> 
>>> On 7/1/2016 10:24 AM, kirk at kodewerk.com wrote:
>>>> Hi Jon,
>>>> 
>>>> I’ve reversed engineered startup parameters and the result is a chart of all possible flag settings and their outcomes. I’d be happy to share it with you as I believe it shows a way to prevent CMS from being turned on. That said, I would consider this approach to be a hack and I’m not sure it should be used. However it does suggest there is a way to prevent CMS from being activated. BTW, it can’t be turned on after the JVM is initialized.
>>>> 
>>>> Kind regards,
>>>> Kirk Pepperdine
>>>> 
>>>>> On Jul 1, 2016, at 5:12 PM, Jon Masamitsu <jon.masamitsu at oracle.com> wrote:
>>>>> 
>>>>> 
>>>>> 
>>>>> On 7/1/2016 7:39 AM, Andrew Haley wrote:
>>>>>> On 01/07/16 15:36, Jon Masamitsu wrote:
>>>>>>> Do you have a suggestion on how continued support for CMS in OpenJDK
>>>>>>> would work?  I've been told that if CMS is built in the binaries that Oracle
>>>>>>> distributes, then customers will expect Oracle to support it. Regardless
>>>>>>> of what the release notes say.
>>>>>> How are they to know that it's in there?
>>>>> If nothing else, it's open source.  :-)
>>>>>>  All you have to do is
>>>>>> disable the option in Oracle builds.
>>>>> It seems like that should be enough, but I honestly don't know.
>>>>> If there's some squirrely way for a customer to turn it on
>>>>> (jcmd, start up script, who knows what), we'd be stuck.
>>>>> I think Oracle is very care about such things.  If asked, "Is there
>>>>> any possible way that CMS could be turned on?",  the best
>>>>> I could do would be to say, "I don't think so".  Not good enough.
>>>>> 
>>>>> Jon
>>>>> 
>>>>> 
>>>>> 
>>>>>> Andrew.
>>>>>> 
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 496 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <http://mail.openjdk.java.net/pipermail/hotspot-gc-dev/attachments/20160702/52f10f66/signature.asc>


More information about the hotspot-gc-dev mailing list