Ian:Another important aspect is the client itself. We usually tend to overlook the cost because we are only thinking of SOLR. The client always has to parse the xml and this cost is usually significant . The same math Yonik has given for server applies to client as well .A typical tomcat usually emits more than 100 req/sec. Assuming that each of these requests fire a SOLR request we are parsing as many xmls and those CPU cycles could have been used for increasing the throughput of the server .
The only benefit of xml is that it is easy to debug. In most of the cases we are debugging the results. Assuming that marshalling and unmarshalling code is bug free we can easily switch to xml mode while debugging and can switch to binary after that.As On Tue, Apr 15, 2008 at 7:05 PM, Yonik Seeley (JIRA) <[EMAIL PROTECTED]> wrote: > > > [ > https://issues.apache.org/jira/browse/SOLR-486?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12589074#action_12589074 > ] > > Yonik Seeley commented on SOLR-486: > ----------------------------------- > > Ian: you can sort of figure out what the total overhead would be given your > max query rate. > For example, XML had an encode rate (for this particular test) of 627 > messages per second (on my 3GHz P4). > If you wanted 100qps, XML encoding would be taking up 16% (100/627) > > Note that the overhead is particularly important for distributed search, > given the number of messages that must be sent+received, and also the limited > network bandwidth it must be done in. > > > > Support binary formats for QueryresponseWriter > > ---------------------------------------------- > > > > Key: SOLR-486 > > URL: https://issues.apache.org/jira/browse/SOLR-486 > > Project: Solr > > Issue Type: Improvement > > Components: clients - java, search > > Reporter: Noble Paul > > Assignee: Yonik Seeley > > Fix For: 1.3 > > > > Attachments: SOLR-486.patch, SOLR-486.patch, SOLR-486.patch, > SOLR-486.patch, SOLR-486.patch, SOLR-486.patch, SOLR-486.patch > > > > > > QueryResponse writer only allows text data to be written. > > So it is not possible to implement a binary protocol . Create another > interface which has a method > > write(OutputStream os, SolrQueryRequest request, SolrQueryResponse > response) > > -- > This message is automatically generated by JIRA. > - > You can reply to this email to add a comment to the issue online. > >
