comments about the 120samples const and link to Gorilla papers. (#423)

Signed-off-by: Krasi Georgiev <kgeorgie@redhat.com>

README.md

@@ -7,6 +7,8 @@ This repository contains the Prometheus storage layer that is used in its 2.x re
 
 A writeup of its design can be found [here](https://fabxc.org/blog/2017-04-10-writing-a-tsdb/).
 
+Based on the Gorilla TSDB [white papers](http://www.vldb.org/pvldb/vol8/p1816-teller.pdf).
+
 Video: [Storing 16 Bytes at Scale](https://youtu.be/b_pEevMAC3I) from [PromCon 2017](https://promcon.io/2017-munich/).
 
 See also the [format documentation](docs/format/README.md).

head.go

@@ -1427,6 +1427,9 @@ func (s *memSeries) truncateChunksBefore(mint int64) (removed int) {
 
 // append adds the sample (t, v) to the series.
 func (s *memSeries) append(t int64, v float64) (success, chunkCreated bool) {
+	// Based on Gorilla white papers this offers near-optimal compression ratio
+	// so anything bigger that this has diminishing returns and increases
+	// the time range within which we have to decompress all samples.
 	const samplesPerChunk = 120
 
 	c := s.head()