Initially, two capped collections were created to implement "cache" feature.

db.createCollection("purchase",

{capped:true, size:10000000}

)
db.purchase.ensureIndex(

{key:1, ts:-1}

,

{background:true}

);
db.createCollection("refinance",

{capped:true, size:10000000}

)
db.refinance.ensureIndex(

{key:1, ts: -1}

,

{background:true}

);

The lookup query is done via

Date now = new Date();
BasicDBObject dbQuery = new BasicDBObject();
dbQuery.put(KEY_FIELD_NAME, requestSignature);
dbQuery.put(TIMESTAMP_FIELD_NAME, new BasicDBObject("$gt",
now.getTime() - query_offer_cache_expiration));

cur = coll.find(dbQuery).addOption(
Bytes.QUERYOPTION_SLAVEOK).sort(
getMongoDBSortCriteria()).limit(1);

We've three servers, each has 24 CPUs w/ total 48G.

We see the query performance seems to degrade after we've thousands of documents. To get thing moving again, we dropped capped collections and recreated them. Now stats whos

> db.refinance.stats()
{
"ns" : "mortgagecache.refinance",
"count" : 591,
"size" : 1235880,
"avgObjSize" : 2091.167512690355,
"storageSize" : 10000128,
"numExtents" : 1,
"nindexes" : 1,
"lastExtentSize" : 10000128,
"paddingFactor" : 1,
"flags" : 0,
"totalIndexSize" : 65536,
"indexSizes" :

{ "key_1_ts_-1" : 65536 }

,
"capped" : 1,
"max" : 2147483647,
"ok" : 1
}

Attaching the cache code