Study the update examination more deeply,i logged all "elapsetime" value of Updateresponse, the result list following: It seems that it spent almost 20 ms on adding/updating one document in general, thus, i called which spend less than 20ms on adding one docs as normal log,and the others were "abnormal" logs. i can`t get a correct suit of solr 1.4, i use solr3.2 which has same performance as solr 1.4 during the test. solr3.5 vs solr 3.2 solr3.5 sum of docs:31998 sum of elapsetime:1218344 ms average: 38.0744 ms /doc sum of normal docs:28409 sum of normal elapsetime:442258 average=15.5675 ms/doc normal percentage:28409/31998 = 88.78% abnormal docs: 3590
solr 3.2 sum of docs:31998 sum of elapsetime:852935 ms average:26.6559 ms /doc sum of normal docs:28416 sum of normal elapsetime:443045 average=15.5914 ms/doc normal percentage:28409/31998 = 88.80% abnormal docs: 3160 What can be analyzed from them? B.R. murphy On Fri, Apr 6, 2012 at 10:28 AM, a sd <liurx.cn@gma il.com> wrote: > hi,Erick. > thanks at first. > I had watched the status of JVM at runtime helped by "jconsole" and > "jmap". > 1,When the "Xmx" was not assigned, then, the "Old Gen" area was full whose > size was up to 1.5Gb and whose major content are instances of "String" , > when the whole size of heap was up to the maximum ( about 2GB), the JVM run > gc() ,which wasted the CPU time,then, the performance was degraded sharply, > which was from 100,000 docs per minute to 10,000 docs per minute, as a > examination, i assigned "Xmx=1024m" purposely, the amount was down to 1000 > docs per minute. > 2,When assigned "Xmx=4096m", i found that the "Old Gen" was up to 2.1 GB > and the all size of JVM was up to 3GB, but, the performance with 100,000 > docs per minute can attained. > During all of the test above, i only adjust the setting of client, which > connect to the identical solr server and i empty the "data" directory of > solr home before every test. > By the way, i know the client code was very ugly occupied so many heap > too, but, i wan`t permitted to promote them before i obtain a benchrank > using solrj 3.5 as much as which the old version did using solrj 1.4. > B.R > murphy > > > On Fri, Apr 6, 2012 at 5:54 AM, Erick Erickson <erickerick...@gmail.com>wrote: > >> "What's memory"? Really, how are you measuring it? >> >> If it's virtual, you don't need to worry about it. Is this >> causing you a real problem or are you just nervous about >> the difference? >> >> Best >> Erick >> >> On Wed, Apr 4, 2012 at 11:23 PM, a sd <liurx...@gmail.com> wrote: >> > hi,all. >> > I have write a program which send data to solr using the "update" >> > request handler, when i adopted server & client library ( namely solrj ) >> > with version 4.0 or 3.2, jvm`s heap size was up to 1.0 G about, but >> ,when i >> > transfer the all of them to solr 3.5 ( both server and client libs), the >> > size of heap was top to 3.0G ! There are the same server configuration >> and >> > the identical program. What`s wrong with the new version of solrj 3.5 , >> i >> > had looked the source code, there is no difference between solrj 3.2 and >> > solrj 3.5 where my program may invoke. How can i do to decrease the >> memory >> > cost by solrj 3.5? >> > Any advice will be appreciated! >> > murphy >> > >
