diff --git a/TOPICS/storage-and-discovery/DHTs.md b/TOPICS/storage-and-discovery/DHTs.md
new file mode 100644
index 0000000..b1c8349
--- /dev/null
+++ b/TOPICS/storage-and-discovery/DHTs.md
@@ -0,0 +1,11 @@
+## Decentralized Hash Tables
+
+### Bigtable as a DHT
+
+It is layers. There is a DHT with 64 megabyte blocks. The tabular data is in column store format in a LSM tree. 
+With a 2^21 fanout the blocks are never more than 3 deep. All nodes have the root. So never more then 2 network hops. 
+They keep the 2nd layer in ram so no more than 1 seek per block. 
+
+And since each block is amortised across the many rows in a column to reach 64 MiB compressed it is very fast at table scans.
+
+This is very different then a ipfs dht with far less fanout and one object retrieved per DHT lookup with poor amortiseation.