lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chris Hostetter <>
Subject Re: Inconsistent response time
Date Fri, 03 Oct 2014 22:21:22 GMT
: Solr for large and time consuming queries. We have found a very inconsistent
: result in the time elapsed when pinging Solr. If we ping Solr from a desktop
: Windows 7 machine, there is usually a 5 ms elapsed time. But if we ping the
: same Solr instance from a Windows Server 2008 machine, it takes about 15 ms.

: Does anybody have any ideas about why this might be? We get these same
: results pretty consistently (testing on multiple desktops and servers). One

the devil is really in the details here ... and you haven't provided very 

Define "multiple desktops and servers" ... do you specifically mean 
multiple *local* desktop machines as the client and multiple *remote* 
servers running solr, or are you also refering to "windows server as 
client talking solr server" situations? ("server is vague and ambiguious 
in your email, particularly since one of hte variables is "Windows Server 

what is your net architecture?

how/where are all of the various machines physically located/connected on 
the network?

Hypothetical cause of your problem that wold fit all of the info you have 
provided: your solr server is a test machine sitting hte same building as 
all of your "desktop" Window 7 machines, and so it's just a local ethernet 
hop away, but your "Windows Server 2008" machine is a "server" in some 
remote data center and has more network hops to go through.

see what i mean about hte details mattering?

when you run this code, is the command line arg you specify an IP addr o a 
hostname? have you ruled out DNS lookup discrepencies between the diff 
client operating systems?

: thing that seemed to have an impact is removing various additional JDKs that
: had been installed, and JDK 1.7u67 specifically seemed to make a difference.

impact how?  Does JDK 1.7u67 make your tests go faster or slower? faster 
or slower then what? what other java versions did you try?

: Finally, the code we are suing to test this is below. If there is a better
: test I would be curious to hear that as well.

well, since the problem you are describing doesn't seem to have anything 
to do with solr, removing solr & solrj completely from the equation would 
be the first thing i would test.

do you have a "curl" equivilent on these windows machines?  have you tried 
just fetching the ping URL w/o using the SolrJ code and comparing hte 
response times that way?

the XML format of the ping response is also really trivial -- you could 
save it to a plain XML file, toss it up in a directory on your favorite 
webserver, and point either curl (or the solrj SolrServer class) at that 
to sanity check wether anything in Solr (or the solrj SolrServer class) is 
actually causing the discrepency you are seeing.

My best guess is the discrepencies you are seeing are entirely network/os 
based and have nothing to do with solr.


View raw message