CPU utilization keeps increasing when using HDFS
Hi We have written a MapReduce application based on Hadoop 2.4 which keeps reading data from HDFS(Pseudo-distributed mode in one node). And we found the CPU system time and user time of the application keeps increasing when it is running. If we changed the application to read data from local disk without changing any other business logic, the CPU utilization kept stable. So we have conclusion that the CPU utilization is related to HDFS. We want to know whether this issue is really related to HDFS and is there any solution to fix it? [cid:image001.png@01CFC5EE.E02C6D50] Thanks a lot! BR/Shiyuan
Re: CPU utilization keeps increasing when using HDFS
Would you please give the output of the top command? at least to show that the HDFS process did use that much of CPU; On Mon, Sep 1, 2014 at 2:19 PM, Shiyuan Xiao shiyuan.x...@ericsson.com wrote: Hi We have written a MapReduce application based on Hadoop 2.4 which keeps reading data from HDFS(Pseudo-distributed mode in one node). And we found the CPU system time and user time of the application keeps increasing when it is running. If we changed the application to read data from local disk without changing any other business logic, the CPU utilization kept stable. So we have conclusion that the CPU utilization is related to HDFS. We want to know whether this issue is really related to HDFS and is there any solution to fix it? Thanks a lot! BR/Shiyuan -- Regards, *Stanley Shi,*
RE: CPU utilization keeps increasing when using HDFS
Because we are running the application with accessing local disk now, I can’t give the “top” command’s output when running with HDFS. But we used “top” and “pidstat” to check the CPU utilization of our application, I can confirm the CPU utilization of our application was increasing and the CPU utilization of datanode, namenode, resourcemanager and NodeManager processes kept stable. Below the “top” command’s output when the application is accessing local disk: [reporting@ms1 ~]$ top top - 15:04:58 up 33 days, 24 min, 3 users, load average: 4.05, 4.08, 3.92 Tasks: 361 total, 1 running, 360 sleeping, 0 stopped, 0 zombie Cpu(s): 34.5%us, 2.3%sy, 0.0%ni, 63.2%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st Mem: 66068256k total, 54013596k used, 12054660k free, 3400140k buffers Swap: 2097144k total, 268376k used, 1828768k free, 41202752k cached PID USER PR NI VIRT RES SHR S %CPU %MEMTIME+ COMMAND 33364 reportin 20 0 1628m 745m 17m S 168.2 1.2 0:05.07 /usr/java/default/bin/java -Djava.net.preferIPv4Stack=true -Dhadoop.me 33130 reportin 20 0 1078m 246m 18m S 130.7 0.4 0:08.10 /usr/java/default/bin/java -Xmx1000m -Djava.library.path=/opt/hadoop/d 33439 reportin 20 0 1613m 143m 17m S 108.1 0.2 0:03.26 /usr/java/default/bin/java -Djava.net.preferIPv4Stack=true -Dhadoop.me 25690 reportin 20 0 1724m 530m 18m S 8.6 0.8 4:31.44 /usr/java/default/bin/java -Dproc_nodemanager -Xmx1000m -Dhadoop.log.di 32879 reportin 20 0 1679m 370m 18m S 6.6 0.6 0:09.13 /usr/java/default/bin/java -Dlog4j.configuration=container-log4j.proper 32642 reportin 20 0 1662m 372m 18m S 6.0 0.6 0:09.22 /usr/java/default/bin/java -Dlog4j.configuration=container-log4j.proper 25200 reportin 20 0 1639m 326m 18m S 2.0 0.5 0:42.49 /usr/java/default/bin/java -Dproc_datanode -Xmx1000m -Djava.library.pat 25576 reportin 20 0 1804m 400m 18m S 2.0 0.6 0:53.13 /usr/java/default/bin/java -Dproc_resourcemanager -Xmx1000m -Dhadoop.lo 25058 reportin 20 0 1622m 401m 18m S 1.7 0.6 0:42.88 /usr/java/default/bin/java -Dproc_namenode -Xmx1000m -Djava.library.pat 33262 reportin 20 0 15260 1556 1012 R 0.7 0.0 0:00.04 top 2984 root 20 0 1227m 14m 1324 S 0.3 0.0 52:00.05 /usr/bin/python /opt/ericsson/nms/litp//bin/landscape_service.py --daem 32019 reportin 20 0 1090m 248m 18m S 0.3 0.4 0:09.21 /usr/java/default/bin/java -Xmx1000m -Djava.library.path=/opt/hadoop/de 49403 hyperic 20 0 5333m 216m 15m S 0.3 0.3 37:31.53 /usr/java/default/jre//bin/java -Djava.security.auth.login.config=../.. 50715 reportin 20 0 5472m 380m 13m S 0.3 0.6 3:57.15 java -Xmx2048m -XX:MaxPermSize=128m -Dlogback.configurationFile=/opt/er 1 root 20 0 19228 1100 896 S 0.0 0.0 10:53.91 /sbin/init 2 root 20 0 000 S 0.0 0.0 0:00.02 [kthreadd] 3 root RT 0 000 S 0.0 0.0 0:27.01 [migration/0] From: Stanley Shi [mailto:s...@pivotal.io] Sent: 2014年9月1日 14:32 To: user@hadoop.apache.org Subject: Re: CPU utilization keeps increasing when using HDFS Would you please give the output of the top command? at least to show that the HDFS process did use that much of CPU; On Mon, Sep 1, 2014 at 2:19 PM, Shiyuan Xiao shiyuan.x...@ericsson.commailto:shiyuan.x...@ericsson.com wrote: Hi We have written a MapReduce application based on Hadoop 2.4 which keeps reading data from HDFS(Pseudo-distributed mode in one node). And we found the CPU system time and user time of the application keeps increasing when it is running. If we changed the application to read data from local disk without changing any other business logic, the CPU utilization kept stable. So we have conclusion that the CPU utilization is related to HDFS. We want to know whether this issue is really related to HDFS and is there any solution to fix it? [cid:image001.png@01CFC5F2.71B29440] Thanks a lot! BR/Shiyuan -- Regards, Stanley Shi, [http://www.gopivotal.com/files/media/logos/pivotal-logo-email-signature.png]
Re: CPU utilization keeps increasing when using HDFS
Because you are using one node Pseudo cluster. When HDFS client write data to HDFS, client will compute the data chunk checksum and the datanode will verify it. It costs cpu shares. You can monitoring the cpu usages for each process. I guess the NameNode cpu usage is OK. But the client process and DataNode process might use most of the cpu shares. On Mon, Sep 1, 2014 at 3:09 PM, Shiyuan Xiao shiyuan.x...@ericsson.com wrote: Because we are running the application with accessing local disk now, I can’t give the “top” command’s output when running with HDFS. But we used “top” and “pidstat” to check the CPU utilization of our application, I can confirm the CPU utilization of our application was increasing and the CPU utilization of datanode, namenode, resourcemanager and NodeManager processes kept stable. Below the “top” command’s output when the application is accessing local disk: [reporting@ms1 ~]$ top top - 15:04:58 up 33 days, 24 min, 3 users, load average: 4.05, 4.08, 3.92 Tasks: 361 total, 1 running, 360 sleeping, 0 stopped, 0 zombie Cpu(s): 34.5%us, 2.3%sy, 0.0%ni, 63.2%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st Mem: 66068256k total, 54013596k used, 12054660k free, 3400140k buffers Swap: 2097144k total, 268376k used, 1828768k free, 41202752k cached PID USER PR NI VIRT RES SHR S %CPU %MEMTIME+ COMMAND 33364 reportin 20 0 1628m 745m 17m S 168.2 1.2 0:05.07 /usr/java/default/bin/java -Djava.net.preferIPv4Stack=true -Dhadoop.me 33130 reportin 20 0 1078m 246m 18m S 130.7 0.4 0:08.10 /usr/java/default/bin/java -Xmx1000m -Djava.library.path=/opt/hadoop/d 33439 reportin 20 0 1613m 143m 17m S 108.1 0.2 0:03.26 /usr/java/default/bin/java -Djava.net.preferIPv4Stack=true -Dhadoop.me 25690 reportin 20 0 1724m 530m 18m S 8.6 0.8 4:31.44 /usr/java/default/bin/java -Dproc_nodemanager -Xmx1000m -Dhadoop.log.di 32879 reportin 20 0 1679m 370m 18m S 6.6 0.6 0:09.13 /usr/java/default/bin/java -Dlog4j.configuration=container-log4j.proper 32642 reportin 20 0 1662m 372m 18m S 6.0 0.6 0:09.22 /usr/java/default/bin/java -Dlog4j.configuration=container-log4j.proper 25200 reportin 20 0 1639m 326m 18m S 2.0 0.5 0:42.49 /usr/java/default/bin/java -Dproc_datanode -Xmx1000m -Djava.library.pat 25576 reportin 20 0 1804m 400m 18m S 2.0 0.6 0:53.13 /usr/java/default/bin/java -Dproc_resourcemanager -Xmx1000m -Dhadoop.lo 25058 reportin 20 0 1622m 401m 18m S 1.7 0.6 0:42.88 /usr/java/default/bin/java -Dproc_namenode -Xmx1000m -Djava.library.pat 33262 reportin 20 0 15260 1556 1012 R 0.7 0.0 0:00.04 top 2984 root 20 0 1227m 14m 1324 S 0.3 0.0 52:00.05 /usr/bin/python /opt/ericsson/nms/litp//bin/landscape_service.py --daem 32019 reportin 20 0 1090m 248m 18m S 0.3 0.4 0:09.21 /usr/java/default/bin/java -Xmx1000m -Djava.library.path=/opt/hadoop/de 49403 hyperic 20 0 5333m 216m 15m S 0.3 0.3 37:31.53 /usr/java/default/jre//bin/java -Djava.security.auth.login.config=../.. 50715 reportin 20 0 5472m 380m 13m S 0.3 0.6 3:57.15 java -Xmx2048m -XX:MaxPermSize=128m -Dlogback.configurationFile=/opt/er 1 root 20 0 19228 1100 896 S 0.0 0.0 10:53.91 /sbin/init 2 root 20 0 000 S 0.0 0.0 0:00.02 [kthreadd] 3 root RT 0 000 S 0.0 0.0 0:27.01 [migration/0] *From:* Stanley Shi [mailto:s...@pivotal.io] *Sent:* 2014年9月1日 14:32 *To:* user@hadoop.apache.org *Subject:* Re: CPU utilization keeps increasing when using HDFS Would you please give the output of the top command? at least to show that the HDFS process did use that much of CPU; On Mon, Sep 1, 2014 at 2:19 PM, Shiyuan Xiao shiyuan.x...@ericsson.com wrote: Hi We have written a MapReduce application based on Hadoop 2.4 which keeps reading data from HDFS(Pseudo-distributed mode in one node). And we found the CPU system time and user time of the application keeps increasing when it is running. If we changed the application to read data from local disk without changing any other business logic, the CPU utilization kept stable. So we have conclusion that the CPU utilization is related to HDFS. We want to know whether this issue is really related to HDFS and is there any solution to fix it? Thanks a lot! BR/Shiyuan -- Regards, *Stanley Shi,* -- Regards Gordon Wang
RE: CPU utilization keeps increasing when using HDFS
Yes, the client process used the most CPU shares. But could you please help explain why the CPU utilization kept increasing? We are sure that the traffic of provisioned data into HDFS was stable. Thanks BR/Shiyuan From: Gordon Wang [mailto:gw...@pivotal.io] Sent: 2014年9月1日 15:48 To: user@hadoop.apache.org Subject: Re: CPU utilization keeps increasing when using HDFS Because you are using one node Pseudo cluster. When HDFS client write data to HDFS, client will compute the data chunk checksum and the datanode will verify it. It costs cpu shares. You can monitoring the cpu usages for each process. I guess the NameNode cpu usage is OK. But the client process and DataNode process might use most of the cpu shares. On Mon, Sep 1, 2014 at 3:09 PM, Shiyuan Xiao shiyuan.x...@ericsson.commailto:shiyuan.x...@ericsson.com wrote: Because we are running the application with accessing local disk now, I can’t give the “top” command’s output when running with HDFS. But we used “top” and “pidstat” to check the CPU utilization of our application, I can confirm the CPU utilization of our application was increasing and the CPU utilization of datanode, namenode, resourcemanager and NodeManager processes kept stable. Below the “top” command’s output when the application is accessing local disk: [reporting@ms1 ~]$ top top - 15:04:58 up 33 days, 24 min, 3 users, load average: 4.05, 4.08, 3.92 Tasks: 361 total, 1 running, 360 sleeping, 0 stopped, 0 zombie Cpu(s): 34.5%us, 2.3%sy, 0.0%ni, 63.2%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st Mem: 66068256k total, 54013596k used, 12054660k free, 3400140k buffers Swap: 2097144k total, 268376k used, 1828768k free, 41202752k cached PID USER PR NI VIRT RES SHR S %CPU %MEMTIME+ COMMAND 33364 reportin 20 0 1628m 745m 17m S 168.2 1.2 0:05.07 /usr/java/default/bin/java -Djava.net.preferIPv4Stack=true -Dhadoop.me 33130 reportin 20 0 1078m 246m 18m S 130.7 0.4 0:08.10 /usr/java/default/bin/java -Xmx1000m -Djava.library.path=/opt/hadoop/d 33439 reportin 20 0 1613m 143m 17m S 108.1 0.2 0:03.26 /usr/java/default/bin/java -Djava.net.preferIPv4Stack=true -Dhadoop.me 25690 reportin 20 0 1724m 530m 18m S 8.6 0.8 4:31.44 /usr/java/default/bin/java -Dproc_nodemanager -Xmx1000m -Dhadoop.log.di 32879 reportin 20 0 1679m 370m 18m S 6.6 0.6 0:09.13 /usr/java/default/bin/java -Dlog4j.configuration=container-log4j.proper 32642 reportin 20 0 1662m 372m 18m S 6.0 0.6 0:09.22 /usr/java/default/bin/java -Dlog4j.configuration=container-log4j.proper 25200 reportin 20 0 1639m 326m 18m S 2.0 0.5 0:42.49 /usr/java/default/bin/java -Dproc_datanode -Xmx1000m -Djava.library.pat 25576 reportin 20 0 1804m 400m 18m S 2.0 0.6 0:53.13 /usr/java/default/bin/java -Dproc_resourcemanager -Xmx1000m -Dhadoop.lo 25058 reportin 20 0 1622m 401m 18m S 1.7 0.6 0:42.88 /usr/java/default/bin/java -Dproc_namenode -Xmx1000m -Djava.library.pat 33262 reportin 20 0 15260 1556 1012 R 0.7 0.0 0:00.04 top 2984 root 20 0 1227m 14m 1324 S 0.3 0.0 52:00.05 /usr/bin/python /opt/ericsson/nms/litp//bin/landscape_service.py --daem 32019 reportin 20 0 1090m 248m 18m S 0.3 0.4 0:09.21 /usr/java/default/bin/java -Xmx1000m -Djava.library.path=/opt/hadoop/de 49403 hyperic 20 0 5333m 216m 15m S 0.3 0.3 37:31.53 /usr/java/default/jre//bin/java -Djava.security.auth.login.config=../.. 50715 reportin 20 0 5472m 380m 13m S 0.3 0.6 3:57.15 java -Xmx2048m -XX:MaxPermSize=128m -Dlogback.configurationFile=/opt/er 1 root 20 0 19228 1100 896 S 0.0 0.0 10:53.91 /sbin/init 2 root 20 0 000 S 0.0 0.0 0:00.02 [kthreadd] 3 root RT 0 000 S 0.0 0.0 0:27.01 [migration/0] From: Stanley Shi [mailto:s...@pivotal.iomailto:s...@pivotal.io] Sent: 2014年9月1日 14:32 To: user@hadoop.apache.orgmailto:user@hadoop.apache.org Subject: Re: CPU utilization keeps increasing when using HDFS Would you please give the output of the top command? at least to show that the HDFS process did use that much of CPU; On Mon, Sep 1, 2014 at 2:19 PM, Shiyuan Xiao shiyuan.x...@ericsson.commailto:shiyuan.x...@ericsson.com wrote: Hi We have written a MapReduce application based on Hadoop 2.4 which keeps reading data from HDFS(Pseudo-distributed mode in one node). And we found the CPU system time and user time of the application keeps increasing when it is running. If we changed the application to read data from local disk without changing any other business logic, the CPU utilization kept stable. So we have conclusion that the CPU utilization is related to HDFS. We want to know whether this issue is really related to HDFS and is there any solution to fix it? [cid:image001.png@01CFC608.634A1070] Thanks a lot! BR/Shiyuan -- Regards, Stanley Shi, [http://www.gopivotal.com/files/media/logos/pivotal-logo-email-signature.png] -- Regards Gordon Wang
