incubator-kato-spec mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Christian Glatschke <christ...@glatschke.com>
Subject Re: Trend Analyser Tool - what should it be able to do?
Date Tue, 27 Jan 2009 10:32:06 GMT
Hi all,

here are some thoughts from my site:

- Which objects have maximum instances across the two dumps?

- What percentage of the heap is occupied by objects of a particular package
across the two dumps ?
- How have the hot spots changed?

For the user story when we have multiple snapshots of the same system and we
want to analyze:

- how is the heap growing
- how frequently is gc happening ?

All this metrics can be delivered easily from within a live system by using the memory metrics
of the jvm.
The only concern might be the cpu overhead, but with a good configuration I can imagine this
could be done.
All this data needs to be stored for historical reviews to analize after an "Out-of-memory"
app crash what caused the memory leak.

This can be real "killer" feature.

Cheers,

Christian Glatschke





Sonal Goyal schrieb:
> Hi Steve,
> I am thinking on the lines of:
>
> - Which objects have maximum instances across the two dumps?
> - What percentage of the heap is occupied by objects of a particular package
> across the two dumps ?
> - How have the hot spots changed?
>
> For the user story when we have multiple snapshots of the same system and we
> want to analyze:
>
> - how is the heap growing
> - how frequently is gc happening ?
> Thanks and Regards,
> Sonal
>
>
> On Thu, Jan 22, 2009 at 2:11 PM, Steve Poole <spoole167@googlemail.com>wrote:
>
>   
>> So what data would you want to compare specifically?
>>
>> On Thu, Jan 22, 2009 at 4:44 AM, Sonal Goyal <sonalgoyal4@gmail.com>
>> wrote:
>>
>>     
>>> Hi,
>>> I think my main user story behind this tool was:
>>>
>>> We release a critical patch onto production and we want to ensure that
>>>       
>> the
>>     
>>> system performance is not impacted. If we could compare the pre and post
>>> patch systems, we would know where things are.
>>>
>>> Thanks and Regards,
>>> Sonal
>>>
>>>
>>> On Wed, Jan 21, 2009 at 8:17 PM, Steve Poole <spoole167@googlemail.com
>>>       
>>>> wrote:
>>>>         
>>>> Hi,  more questions for the Expert Group :-)
>>>>
>>>> The trend analysis tool concept we've identified has no substance
>>>>         
>> behind
>>     
>>> it
>>>       
>>>> yet.
>>>>
>>>> The reason this tool is on the list is that we identified that there
>>>>         
>> were
>>     
>>>> situations where the monitoring of live systems was not effective  -
>>>>         
>> the
>>     
>>>> main reason cited was that  the live system was changing too much for
>>>>         
>> the
>>     
>>>> monitoring system to cope with . It was felt that the snapshot approach
>>>> would allow the monitoring system to provide more accurate trending at
>>>>         
>>> the
>>>       
>>>> cost of it being less frequent.
>>>>
>>>> Its easy to say "lets show trends between snapshots" but what does that
>>>> really mean?
>>>>
>>>>         
>
>   


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message