Hadoop上的Data Locality是指數(shù)據(jù)與Mapper任務(wù)運(yùn)行時(shí)數(shù)據(jù)的距離接近程度(Data Locality in Hadoop refers to the“proximity” of the data with respect to the Mapper tasks working on the data.)
1. why data locality is imporant?
當(dāng)數(shù)據(jù)集存儲(chǔ)在HDFS中時(shí),它被劃分為塊并存儲(chǔ)在Hadoop集群中的DataNode上。當(dāng)在數(shù)據(jù)集執(zhí)行MapReduce作業(yè)時(shí),各個(gè)Mappers將處理這些塊(輸進(jìn)行入分片處理)。如果Mapper不能從它執(zhí)行的節(jié)點(diǎn)上獲取數(shù)據(jù),數(shù)據(jù)需要通過(guò)網(wǎng)絡(luò)從具有這些數(shù)據(jù)的DataNode拷貝到執(zhí)行Mapper任務(wù)的節(jié)點(diǎn)上(the data needs to be copied over the network from the DataNode which has the data to the DataNode which is executing the Mapper task)。假設(shè)一個(gè)MapReduce作業(yè)具有超過(guò)1000個(gè)Mapper,在同一時(shí)間每一個(gè)Mapper都試著去從集群上另一個(gè)DataNode節(jié)點(diǎn)上拷貝數(shù)據(jù),這將導(dǎo)致嚴(yán)重的網(wǎng)絡(luò)阻塞,因?yàn)樗械腗apper都嘗試在同一時(shí)間拷貝數(shù)據(jù)(這不是一種理想的方法)。因此,將計(jì)算任務(wù)移動(dòng)到更接近數(shù)據(jù)的節(jié)點(diǎn)上是一種更有效與廉價(jià)的方法,相比于將數(shù)據(jù)移動(dòng)到更接近計(jì)算任務(wù)的節(jié)點(diǎn)上(it is always effective and cheap to move the computation closer to the data than to move the data closer to the computation)。
2. How is data proximity defined?
當(dāng)JobTracker(MRv1)或ApplicationMaster(MRv2)接收到運(yùn)行作業(yè)的請(qǐng)求時(shí),它查看集群中的哪些節(jié)點(diǎn)有足夠的資源來(lái)執(zhí)行該作業(yè)的Mappers和Reducers。同時(shí)需要根據(jù)Mapper運(yùn)行數(shù)據(jù)所處位置來(lái)考慮決定每個(gè)Mapper執(zhí)行的節(jié)點(diǎn)(serious consideration is made to decide on which nodes the individual Mappers will be executed based on where the data for the Mapper is located)。
3. Data Local
當(dāng)數(shù)據(jù)所處的節(jié)點(diǎn)與Mapper執(zhí)行的節(jié)點(diǎn)是同一節(jié)點(diǎn),我們稱之為Data Local。在這種情況下,數(shù)據(jù)的接近度更接近計(jì)算( In this case the proximity of the data is closer to the computation.)。JobTracker(MRv1)或ApplicationMaster(MRv2)首選具有Mapper所需要數(shù)據(jù)的節(jié)點(diǎn)來(lái)執(zhí)行Mapper。
4. Rack Local
雖然Data Local是理想的選擇,但由于受限于集群上的資源,并不總是在與數(shù)據(jù)同一節(jié)點(diǎn)上執(zhí)行Mapper(Although Data Local is the ideal choice, it is not always possible to execute the Mapper on the same node as the data due to resource constraints on a busy cluster)。在這種情況下,優(yōu)選地選擇在那些與數(shù)據(jù)節(jié)點(diǎn)在同一機(jī)架上的不同節(jié)點(diǎn)上運(yùn)行Mapper( In such instances it is preferred to run the Mapper on a different node but on the same rack as the node which has the data.)。在這種情況下,數(shù)據(jù)將在節(jié)點(diǎn)之間進(jìn)行移動(dòng),從具有數(shù)據(jù)的節(jié)點(diǎn)移動(dòng)到在同一機(jī)架上執(zhí)行Mapper的節(jié)點(diǎn),這種情況我們稱之為Rack Local。
5. Different Rack
在繁忙的群集中,有時(shí)Rack Local也不可能。在這種情況下,選擇不同機(jī)架上的節(jié)點(diǎn)來(lái)執(zhí)行Mapper,并且將數(shù)據(jù)從具有數(shù)據(jù)的節(jié)點(diǎn)復(fù)制到在不同機(jī)架上執(zhí)行Mapper的節(jié)點(diǎn)。這是最不可取的情況。
如有疑問(wèn)請(qǐng)留言或者到本站社區(qū)交流討論,感謝閱讀,希望能幫助到大家,謝謝大家對(duì)本站的支持!
原文鏈接:http://blog.csdn.net/zhyooo123/article/details/77868170