Hi All,
I see that we have two executors on master, but the machine has only one CPU. This wouldn't be a problem in general - it possibly gives some speedup when one of the builds is waiting for some external resource - but it adds way too much noise to the SqueakTrunkPerformance job's results. The last few builds are quite okay, but previous runs look like noise only: http://build.squeak.org/job/SqueakTrunkPerformance/performance/
So, I think we should either ensure somehow that the SqueakTrunkPerformance job can only be executed alone, or just decrease the number of executors to one. Another thing which should be ensured is that this job can't be run on any of the slaves, because that makes the comparision with previous builds uncomparable.
I can imagine that these are all handled already, because the last few results are pretty much consistent. In that case sorry for the noise.
Levente
On 10.04.2015, at 17:38, Levente Uzonyi leves@elte.hu wrote:
Hi All,
I see that we have two executors on master, but the machine has only one CPU. This wouldn't be a problem in general - it possibly gives some speedup when one of the builds is waiting for some external resource - but it adds way too much noise to the SqueakTrunkPerformance job's results. The last few builds are quite okay, but previous runs look like noise only: http://build.squeak.org/job/SqueakTrunkPerformance/performance/
So, I think we should either ensure somehow that the SqueakTrunkPerformance job can only be executed alone, or just decrease the number of executors to one. Another thing which should be ensured is that this job can't be run on any of the slaves, because that makes the comparision with previous builds uncomparable.
I can imagine that these are all handled already, because the last few results are pretty much consistent. In that case sorry for the noise.
Another way is to ensure it runs on another host which provides just one executor. best -tobias
On Fri, 10 Apr 2015, Tobias Pape wrote:
On 10.04.2015, at 17:38, Levente Uzonyi leves@elte.hu wrote:
Hi All,
I see that we have two executors on master, but the machine has only one CPU. This wouldn't be a problem in general - it possibly gives some speedup when one of the builds is waiting for some external resource - but it adds way too much noise to the SqueakTrunkPerformance job's results. The last few builds are quite okay, but previous runs look like noise only: http://build.squeak.org/job/SqueakTrunkPerformance/performance/
So, I think we should either ensure somehow that the SqueakTrunkPerformance job can only be executed alone, or just decrease the number of executors to one. Another thing which should be ensured is that this job can't be run on any of the slaves, because that makes the comparision with previous builds uncomparable.
I can imagine that these are all handled already, because the last few results are pretty much consistent. In that case sorry for the noise.
Another way is to ensure it runs on another host which provides just one executor.
But then we need another host. IMHO it's easier to have just one executor on master.
Levente
best -tobias
On 10.04.2015, at 20:30, Levente Uzonyi leves@elte.hu wrote:
On Fri, 10 Apr 2015, Tobias Pape wrote:
On 10.04.2015, at 17:38, Levente Uzonyi leves@elte.hu wrote:
Hi All,
I see that we have two executors on master, but the machine has only one CPU. This wouldn't be a problem in general - it possibly gives some speedup when one of the builds is waiting for some external resource - but it adds way too much noise to the SqueakTrunkPerformance job's results. The last few builds are quite okay, but previous runs look like noise only: http://build.squeak.org/job/SqueakTrunkPerformance/performance/
So, I think we should either ensure somehow that the SqueakTrunkPerformance job can only be executed alone, or just decrease the number of executors to one. Another thing which should be ensured is that this job can't be run on any of the slaves, because that makes the comparision with previous builds uncomparable.
I can imagine that these are all handled already, because the last few results are pretty much consistent. In that case sorry for the noise.
Another way is to ensure it runs on another host which provides just one executor.
But then we need another host. IMHO it's easier to have just one executor on master.
There's mine now. We could introduce a tag 'single' and then everything should run rather automagically best -tobias
Levente
best -tobias
It seems like the SqueakTrunkPerformance job has some problems. The results are consistent, because they are probably not updated at all. The last 29 values are exactly the same for all benchmakrs, for example 9575ms for BinaryTrees: http://build.squeak.org/job/SqueakTrunkPerformance/performance/trendReport/?...
Levete
On Fri, 10 Apr 2015, Levente Uzonyi wrote:
Hi All,
I see that we have two executors on master, but the machine has only one CPU. This wouldn't be a problem in general - it possibly gives some speedup when one of the builds is waiting for some external resource - but it adds way too much noise to the SqueakTrunkPerformance job's results. The last few builds are quite okay, but previous runs look like noise only: http://build.squeak.org/job/SqueakTrunkPerformance/performance/
So, I think we should either ensure somehow that the SqueakTrunkPerformance job can only be executed alone, or just decrease the number of executors to one. Another thing which should be ensured is that this job can't be run on any of the slaves, because that makes the comparision with previous builds uncomparable.
I can imagine that these are all handled already, because the last few results are pretty much consistent. In that case sorry for the noise.
Levente
box-admins@lists.squeakfoundation.org