Just one D has been the plan... On Jun 30, 2011 7:40 AM, "Kelly Stirman" <[email protected]> wrote: > Harry, > > If you have caching (ie, squid), few writes, and only 100K documents, I think you are describing a very modest system, unless you have *lots* of queries per second. > > Most of the work in a request is evaluated on the D node. For example, if you have facets, all the index resolution is performed on the D nodes, with filtering and aggregation on the E node. > > In my experience, most clusters have more D nodes than E nodes. It seems like lately my clusters have had all nodes configured as D/E, but those have all been less than 10 nodes, and the XQuery to be evaluated was only hundreds or maybe thousands of lines for each request. If you have lots of XQuery to be evaluated, you might find more need for E nodes. > > So, how many D nodes were you planning on? > > Kelly > > Message: 2 > Date: Wed, 29 Jun 2011 14:27:17 -0600 > From: Harry Bakken <[email protected]> > Subject: Re: [MarkLogic Dev General] E Node Cluster Question > To: General MarkLogic Developer Discussion > <[email protected]> > Message-ID: <[email protected]> > Content-Type: text/plain; charset="iso-8859-1" > > Hypothetically, I am looking at a system where almost all of our heavy use/traffic is read-only. We'd have applications for managing the content that does the read/write, but write would a very small percentage of the overall use. Document number would be low by ML standards- less than 100K documents. > > > On Wed, Jun 29, 2011 at 2:22 PM, Kelly Stirman > <[email protected]>wrote: > >> Hi Harry, >> >> This begs the question - how many D nodes do you have? How many documents? >> How large is the database? How many reads and writes? >> >> Kelly >> >> Message: 1 >> Date: Wed, 29 Jun 2011 13:26:54 -0600 >> From: Harry Bakken <[email protected]> >> Subject: [MarkLogic Dev General] E Node Cluster Question >> To: General MarkLogic Developer Discussion >> <[email protected]> >> Message-ID: <[email protected]> >> Content-Type: text/plain; charset="iso-8859-1" >> >> I am wondering if anyone can provide real world experience-based >> advice on clustering E nodes. Is it better for system performance to >> run with fewer beefy nodes or more not-so-beefy nodes? For example, >> would it be better to >> have: >> - two E Nodes built on 24 cores and 78 GB of RAM or >> - five E nodes with 8 cores and 16 GB of RAM >> >> Any insight or examples is appreciated. We have some internal info >> from other projects that give us some direction, but I am interested >> to know what others are doing and how MarkLogic is engineered for >> clustering. Is there a performance hit for more nodes? Does it really >> matter? Is it more a question of how much load there is to balance and >> hardware budget than system performance considerations? >> >> Thank you in advance, >> Harry >> _______________________________________________ >> General mailing list >> [email protected] >> http://developer.marklogic.com/mailman/listinfo/general >> > -------------- next part -------------- > An HTML attachment was scrubbed... > URL: http://developer.marklogic.com/pipermail/general/attachments/20110629/78fbbce5/attachment-0001.html > > ------------------------------ > _______________________________________________ > General mailing list > [email protected] > http://developer.marklogic.com/mailman/listinfo/general
_______________________________________________ General mailing list [email protected] http://developer.marklogic.com/mailman/listinfo/general
