incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Samuel CARRIERE <samuel.carri...@urssaf.fr>
Subject Re: Frequent Full GC that take > 30s
Date Tue, 06 May 2014 08:55:17 GMT
>>>Hello.
>>>
>>>I'm having problems with frequent Full GCs that take a long time, and 
cause a 
>>>burst of timeouts for the client application. But first, here is my 
>>>configuration:
>>>
>>>Cassandra: 1.1.5
>>>Key Cache        : size 209715168 (bytes), capacity 209715168 (bytes), 
>>>1331992571 hits, 1831790912 requests, 0.854 recent hit rate, 14400 save 
period 
>>>in seconds
>>>Row Cache        : size 104588313 (bytes), capacity 104857600 (bytes), 
>>>585590776 hits, 646992444 requests, 0.880 recent hit rate, 0 save 
period in 
>>>seconds
>>>flush_largest_memtables_at: 0.8
>>>memtable_total_space_in_mb: 2048
>>>
>>>JVM: 
>>>java version "1.6.0_35"
>>>Java(TM) SE Runtime Environment (build 1.6.0_35-b10)
>>>Java HotSpot(TM) 64-Bit Server VM (build 20.10-b01, mixed mode)
>>>
>>>JVM Options:
>>>-XX:+UseParNewGC
>>>-XX:+UseConcMarkSweepGC
>>>-XX:+CMSParallelRemarkEnabled
>>>-XX:SurvivorRatio=8
>>>-XX:MaxTenuringThreshold=1
>>>-XX:CMSInitiatingOccupancyFraction=75
>>>-XX:+UseCMSInitiatingOccupancyOnly
>>>-XX:ParallelCMSThreads=4
>>>-Xms16G
>>>-Xmx16G
>>>-Xmn800M
>>>-Xss196k
>>>
>>>6 nodes with 32GB RAM, Intel(R) Xeon(R) CPU E5-2609, 300GB data per 
node.

>> Your ParNew size is way too small. Generally 4GB ParNew (-Xmn) works 
out best 
>> for 16GB heap
>
>I was afraid that a 4GB ParNew would cause Young GCs to take too long. 
I'm 
>going to test higher ParNew values.
>
>Thanks,
>André

Hi André,
What are the results of your investigations ? Did you try to tune 
MaxTenuringThreshold and -Xmn ?
I'm interested because I'm experiencing long GC pauses during repairs 
(cassandra 1.1.12).

Mime
View raw message