incubator-couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paul Carey" <>
Subject grouped results calculated on each query
Date Sun, 30 Nov 2008 16:23:23 GMT

I ran some benchmarks using httperf against a small EC2 instance.

Read performance was roughly the same whether I requested a doc by id
or via a map view. CouchDB consistently served about 100 req/sec.

The number of docs made no difference to the query time for map views,
or reduce views with group=false. Things change as soon as a
group_level is specified.

With 38k key/doc pairs emitted by the map function, response rate
dropped to about 2.5 req/sec when querying with group_level=2. For
1580k docs, it was down to 0.5 req/sec.

Querying the 1580k docs with differing group levels gave me the
following response times

group_level=0 - 0.035s (1 x reduce, 1 x rereduce - 271 bytes)
group_level=1 - 0.120s (1 x reduce, 1 x rereduce - 308 bytes)
group_level=2 - 2.010s (79 x reduce, 79 x rereduce - 26531 bytes)

What I'm wondering is why the per group level results of a reduce
aren't stored in the view index.



View raw message