Why is ascending geo distance sorting faster than descending geo distance sorting - sorting

Question

I'm using Elasticsearch 6.6 and have an index (1 shard, 1 replica) with the geonames (https://www.geonames.org/) dataset indexed (indexsize =1.3 gb, 11.8 mio geopoints).
I was playing around a bit with the geo distance sorting query, sorting the whole index for some origin points. So after some testing I saw that sorting ascending is always faster than sorting descending. here is an example query (i also tested with bigger "size"-parameter):
POST /geonames/_search?request_cache=false
{
"size":1,
"sort" : [
{
"_geo_distance" : {
"location" : [8, 49],
"order" : "asc",
"unit" : "m",
"mode" : "min",
"distance_type" : "arc",
"ignore_unmapped": true
}
}
]
}
Here is the answer for ascending sorting (with explain and profile True):
{
"took" : 1374,
"timed_out" : false,
"_shards" : {
"total" : 1,
"successful" : 1,
"skipped" : 0,
"failed" : 0
},
"hits" : {
"total" : 11858060,
"max_score" : null,
"hits" : [
{
"_shard" : "[geonames][0]",
"_node" : "qXTymyB9QLmxhPtGEtA_mA",
"_index" : "geonames",
"_type" : "doc",
"_id" : "L781LmkBrQo0YN4qP48D",
"_score" : null,
"_source" : {
"id" : "3034701",
"name" : "Forêt de Wissembourg",
"location" : {
"lat" : "49.00924",
"lon" : "8.01542"
}
},
"sort" : [
1523.4121312414704
],
"_explanation" : {
"value" : 1.0,
"description" : "*:*",
"details" : [ ]
}
}
]
},
"profile" : {
"shards" : [
{
"id" : "[qXTymyB9QLmxhPtGEtA_mA][geonames][0]",
"searches" : [
{
"query" : [
{
"type" : "MatchAllDocsQuery",
"description" : "*:*",
"time_in_nanos" : 265223567,
"breakdown" : {
"score" : 0,
"build_scorer_count" : 54,
"match_count" : 0,
"create_weight" : 10209,
"next_doc" : 253091268,
"match" : 0,
"create_weight_count" : 1,
"next_doc_count" : 11858087,
"score_count" : 0,
"build_scorer" : 263948,
"advance" : 0,
"advance_count" : 0
}
}
],
"rewrite_time" : 1097,
"collector" : [
{
"name" : "CancellableCollector",
"reason" : "search_cancelled",
"time_in_nanos" : 1044167746,
"children" : [
{
"name" : "SimpleFieldCollector",
"reason" : "search_top_hits",
"time_in_nanos" : 508296683
}
]
}
]
}
],
"aggregations" : [ ]
}
]
}
}
and here for descending, just switched the parameter from asc to desc (also with profile and explain):
{
"took" : 2226,
"timed_out" : false,
"_shards" : {
"total" : 1,
"successful" : 1,
"skipped" : 0,
"failed" : 0
},
"hits" : {
"total" : 11858060,
"max_score" : null,
"hits" : [
{
"_shard" : "[geonames][0]",
"_node" : "qXTymyB9QLmxhPtGEtA_mA",
"_index" : "geonames",
"_type" : "doc",
"_id" : "Mq80LmkBrQo0YN4q11bA",
"_score" : null,
"_source" : {
"id" : "4036351",
"name" : "Bollons Seamount",
"location" : {
"lat" : "-49.66667",
"lon" : "-176.16667"
}
},
"sort" : [
1.970427111052182E7
],
"_explanation" : {
"value" : 1.0,
"description" : "*:*",
"details" : [ ]
}
}
]
},
"profile" : {
"shards" : [
{
"id" : "[qXTymyB9QLmxhPtGEtA_mA][geonames][0]",
"searches" : [
{
"query" : [
{
"type" : "MatchAllDocsQuery",
"description" : "*:*",
"time_in_nanos" : 268521404,
"breakdown" : {
"score" : 0,
"build_scorer_count" : 54,
"match_count" : 0,
"create_weight" : 9333,
"next_doc" : 256458664,
"match" : 0,
"create_weight_count" : 1,
"next_doc_count" : 11858087,
"score_count" : 0,
"build_scorer" : 195265,
"advance" : 0,
"advance_count" : 0
}
}
],
"rewrite_time" : 1142,
"collector" : [
{
"name" : "CancellableCollector",
"reason" : "search_cancelled",
"time_in_nanos" : 1898324618,
"children" : [
{
"name" : "SimpleFieldCollector",
"reason" : "search_top_hits",
"time_in_nanos" : 1368306442
}
]
}
]
}
],
"aggregations" : [ ]
}
]
}
}
So my question is, why is it like this ? As I understood Es calculates the distance from the origin point to every other point and then sorts them. So why is the descending sorting so much slower ?

Answer 1

Asking the same question on the Elasticsearch board and getting an answer.
So apparantly Elasticsearch uses differnt searching strategies/algorithms for ascending end descending distance sorting.
For the descending sorting it calculates the distance from the origin to every point end then sorts.
For the ascending sorting it uses boundingboxes to filter points near the origin and only calculate the distances for points inside the boundingboxes.