diff --git a/postgresql/postgres.nix b/postgresql/postgres.nix
index 1a39e39..fbbb787 100644
--- a/postgresql/postgres.nix
+++ b/postgresql/postgres.nix
@@ -1,4 +1,27 @@
 {
+  "t3a.nano" = {
+     max_connections = 40;
+     shared_buffers = "128MB";
+     work_mem = "655kB";
+     maintenance_work_mem = "32MB";
+     max_worker_processes = 2;
+     max_parallel_workers_per_gather = 1;
+     max_parallel_workers = 2;
+     max_parallel_maintenance_workers = 1;
+
+     effective_cache_size = "384MB";
+     default_statistics_target = 100;
+     effective_io_concurrency = 200;
+
+     wal_level = "minimal";
+     archive_mode = "off";
+     max_wal_senders = 0;
+     wal_buffers = "3932kB";
+     min_wal_size = "2GB";
+     max_wal_size = "8GB";
+     random_page_cost = 1.1;
+     checkpoint_completion_target = 0.9;
+  };
   "t3a.micro" = {
     max_connections = 60;
     shared_buffers = "256MB";
@@ -43,6 +66,58 @@
     max_wal_size = "4GB";
     min_wal_size = "1GB";
   };
+  "t3a.xlarge" = {
+    max_connections = 200;
+    shared_buffers = "4GB";
+    effective_cache_size = "12GB";
+    maintenance_work_mem = "1GB";
+    checkpoint_completion_target = 0.9;
+    wal_buffers = "16MB";
+    random_page_cost = 1.1;
+    effective_io_concurrency = 200;
+    work_mem = "10485kB";
+    max_worker_processes = 4;
+    max_parallel_workers_per_gather = 2;
+    max_parallel_workers = 4;
+    max_parallel_maintenance_workers = 2;
+  };
+  "t3a.2xlarge" = {
+    max_connections = 200;
+    shared_buffers = "8GB";
+    effective_cache_size = "24GB";
+    maintenance_work_mem = "2GB";
+    checkpoint_completion_target = 0.9;
+    wal_buffers = "16MB";
+    random_page_cost = 1.1;
+    effective_io_concurrency = 200;
+    work_mem = "10485kB";
+    max_worker_processes = 8;
+    max_parallel_workers_per_gather = 4;
+    max_parallel_workers = 8;
+    max_parallel_maintenance_workers = 4;
+  };
+  "m5a.large" = {
+    max_connections = 200;
+    shared_buffers = "2GB";
+    work_mem = "10485kB";
+    maintenance_work_mem = "512MB";
+    max_worker_processes = 2;
+    max_parallel_maintenance_workers = 1;
+    max_parallel_workers_per_gather = 1;
+    max_parallel_workers = 2;
+
+    effective_cache_size = "6GB";
+    default_statistics_target = 100;
+    effective_io_concurrency = 200;
+
+    wal_level = "minimal";
+    archive_mode = "off";
+    max_wal_senders = 0;
+    wal_compression = "on";
+    wal_buffers = "16MB";
+    min_wal_size = "1GB";
+    max_wal_size = "4GB";
+  };
   "m5a.xlarge" = {
     max_connections = 240;
     shared_buffers = "4GB";
@@ -101,6 +176,72 @@
     default_statistics_target = 100;
     effective_io_concurrency = 200;
 
+    wal_level = "minimal";
+    archive_mode = "off";
+    max_wal_senders = 0;
+    wal_compression = "on";
+    wal_buffers = "16MB";
+    max_wal_size = "4GB";
+    min_wal_size = "2GB";
+  };
+  "m5a.8xlarge" = {
+    max_connections = 600;
+    shared_buffers = "32GB";
+    work_mem = "27962kB";
+    maintenance_work_mem = "2GB";
+    max_worker_processes = 32;
+    max_parallel_maintenance_workers = 4;
+    max_parallel_workers_per_gather = 4;
+    max_parallel_workers = 32;
+
+    effective_cache_size = "96GB";
+    default_statistics_target = 100;
+    effective_io_concurrency = 200;
+
+    wal_level = "minimal";
+    archive_mode = "off";
+    max_wal_senders = 0;
+    wal_compression = "on";
+    wal_buffers = "16MB";
+    min_wal_size = "2GB";
+    max_wal_size = "8GB";
+  };
+  "m5a.12xlarge" = {
+    max_connections = 800;
+    shared_buffers = "48GB";
+    work_mem = "41943kB";
+    maintenance_work_mem = "2GB";
+    max_worker_processes = 48;
+    max_parallel_maintenance_workers = 4;
+    max_parallel_workers_per_gather = 4;
+    max_parallel_workers = 48;
+
+    effective_cache_size = "144GB";
+    default_statistics_target = 100;
+    effective_io_concurrency = 200;
+
+    wal_level = "minimal";
+    archive_mode = "off";
+    max_wal_senders = 0;
+    wal_compression = "on";
+    wal_buffers = "16MB";
+    max_wal_size = "4GB";
+    min_wal_size = "2GB";
+  };
+  "m5a.16xlarge" = {
+    max_connections = 1000;
+    shared_buffers = "64GB";
+    work_mem = "83886kB";
+    maintenance_work_mem = "2GB";
+    max_worker_processes = 64;
+    max_parallel_maintenance_workers = 4;
+    max_parallel_workers_per_gather = 4;
+    max_parallel_workers = 64;
+
+    effective_cache_size = "192GB";
+    default_statistics_target = 100;
+    effective_io_concurrency = 200;
+
     wal_level = "minimal";
     archive_mode = "off";
     max_wal_senders = 0;
diff --git a/postgrest/PGBENCH_RESULTS.md b/postgrest/PGBENCH_RESULTS.md
new file mode 100644
index 0000000..f3ae17f
--- /dev/null
+++ b/postgrest/PGBENCH_RESULTS.md
@@ -0,0 +1,211 @@
+ <!--pgrbench-all-pg-instances repeat 3 pgrbench-pgbench -c 100 > PGBENCH_RESULTS.md -->
+
+Running on a m5a.large
+
+
+Run 1:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 45492
+latency average = 66.235 ms
+tps = 1509.766608 (including connections establishing)
+tps = 1510.166881 (excluding connections establishing)
+
+Run 2:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 50828
+latency average = 59.331 ms
+tps = 1685.471195 (including connections establishing)
+tps = 1686.528635 (excluding connections establishing)
+
+Run 3:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 53997
+latency average = 55.817 ms
+tps = 1791.559195 (including connections establishing)
+tps = 1792.247708 (excluding connections establishing)
+
+Running on a m5a.xlarge
+
+
+Run 1:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 92804
+latency average = 32.448 ms
+tps = 3081.890468 (including connections establishing)
+tps = 3082.342735 (excluding connections establishing)
+
+Run 2:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 114382
+latency average = 26.316 ms
+tps = 3799.897694 (including connections establishing)
+tps = 3800.595158 (excluding connections establishing)
+
+Run 3:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 115656
+latency average = 26.068 ms
+tps = 3836.188151 (including connections establishing)
+tps = 3836.812069 (excluding connections establishing)
+
+Running on a m5a.2xlarge
+
+
+Run 1:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 142970
+latency average = 21.036 ms
+tps = 4753.743995 (including connections establishing)
+tps = 4754.249103 (excluding connections establishing)
+
+Run 2:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 233928
+latency average = 12.860 ms
+tps = 7775.855491 (including connections establishing)
+tps = 7776.714704 (excluding connections establishing)
+
+Run 3:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 235969
+latency average = 12.761 ms
+tps = 7836.620631 (including connections establishing)
+tps = 7837.585665 (excluding connections establishing)
+
+Running on a m5a.4xlarge
+
+
+Run 1:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 171468
+latency average = 17.536 ms
+tps = 5702.618662 (including connections establishing)
+tps = 5703.228351 (excluding connections establishing)
+
+Run 2:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 351608
+latency average = 8.555 ms
+tps = 11688.693390 (including connections establishing)
+tps = 11690.031243 (excluding connections establishing)
+
+Run 3:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 355595
+latency average = 8.454 ms
+tps = 11828.751758 (including connections establishing)
+tps = 11830.103914 (excluding connections establishing)
+
+Running on a m5a.8xlarge
+
+
+Run 1:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 176047
+latency average = 17.082 ms
+tps = 5854.134383 (including connections establishing)
+tps = 5854.820377 (excluding connections establishing)
+
+Run 2:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 372931
+latency average = 8.062 ms
+tps = 12403.998820 (including connections establishing)
+tps = 12405.350491 (excluding connections establishing)
+
+Run 3:
+
+transaction type: <builtin: TPC-B (sort of)>
+scaling factor: 50
+query mode: prepared
+number of clients: 100
+number of threads: 8
+duration: 30 s
+number of transactions actually processed: 377333
+latency average = 7.968 ms
+tps = 12550.028498 (including connections establishing)
+tps = 12551.435376 (excluding connections establishing)
diff --git a/postgrest/deploy.nix b/postgrest/deploy.nix
index 71f7f43..b40fcd1 100644
--- a/postgrest/deploy.nix
+++ b/postgrest/deploy.nix
@@ -11,6 +11,7 @@ let
     pgrstPool         = builtins.getEnv "PGRBENCH_PGRST_POOL";
   };
   pkgs = import <nixpkgs> {};
+  postgresConfigs = import ../postgresql/postgres.nix;
 in {
   network.description = "postgrest benchmark";
 
@@ -187,12 +188,13 @@ in {
   client = {nodes, resources, ...}: {
     environment.systemPackages = [
       pkgs.k6
+      pkgs.postgresql_12 # only used for getting pgbench, no postgresql is started here
     ];
     deployment = {
       targetEnv = "ec2";
       ec2 = {
         inherit region accessKeyId;
-        instanceType             = "t3a.xlarge";
+        instanceType             = "t3a.2xlarge";
         associatePublicIpAddress = true;
         keyPair                  = resources.ec2KeyPairs.pgrstBenchKeyPair;
         subnetId                 = resources.vpcSubnets.pgrstBenchSubnet;
@@ -208,12 +210,13 @@ in {
     ];
     networking.hosts = {
       "${nodes.pgrstServer.config.networking.privateIPv4}"   = [ "pgrst" ];
+      "${nodes.pg.config.networking.privateIPv4}"            = [ "pg" ];
     };
   };
 }
 // pkgs.lib.optionalAttrs env.withSeparatePg
 {
-  pg = {resources, ...}: rec {
+  pg = {resources, config, ...}: rec {
     deployment = {
       targetEnv = "ec2";
       ec2 = {
@@ -242,55 +245,24 @@ in {
       '';
       enableTCPIP = true; # listen_adresses = *
       # Tuned according to https://pgtune.leopard.in.ua
-      settings =
-        if deployment.ec2.instanceType == "t3a.nano" then {
-          max_connections = 200;
-          shared_buffers = "128MB";
-          effective_cache_size = "384MB";
-          maintenance_work_mem = "32MB";
-          checkpoint_completion_target = 0.9;
-          wal_buffers = "3932kB";
-          random_page_cost = 1.1;
-          effective_io_concurrency = 200;
-          work_mem = "655kB";
-          max_worker_processes = 2;
-          max_parallel_workers_per_gather = 1;
-          max_parallel_workers = 2;
-          max_parallel_maintenance_workers = 1;
-        }
-        else if deployment.ec2.instanceType == "t3a.xlarge" then {
-          max_connections = 200;
-          shared_buffers = "4GB";
-          effective_cache_size = "12GB";
-          maintenance_work_mem = "1GB";
-          checkpoint_completion_target = 0.9;
-          wal_buffers = "16MB";
-          random_page_cost = 1.1;
-          effective_io_concurrency = 200;
-          work_mem = "10485kB";
-          max_worker_processes = 4;
-          max_parallel_workers_per_gather = 2;
-          max_parallel_workers = 4;
-          max_parallel_maintenance_workers = 2;
-        }
-        else if deployment.ec2.instanceType == "t3a.2xlarge" then {
-          max_connections = 200;
-          shared_buffers = "8GB";
-          effective_cache_size = "24GB";
-          maintenance_work_mem = "2GB";
-          checkpoint_completion_target = 0.9;
-          wal_buffers = "16MB";
-          random_page_cost = 1.1;
-          effective_io_concurrency = 200;
-          work_mem = "10485kB";
-          max_worker_processes = 8;
-          max_parallel_workers_per_gather = 4;
-          max_parallel_workers = 8;
-          max_parallel_maintenance_workers = 4;
-        }
-        else {};
+      settings = builtins.getAttr config.deployment.ec2.instanceType postgresConfigs;
       initialScript = ../schemas/chinook/chinook.sql; # Here goes the sample db
     };
+    # initialize the pgbench db by prepending to the default postgresql systemd post start
+    # https://github.com/NixOS/nixpkgs/blob/master/nixos/modules/services/databases/postgresql.nix#L342-L353
+    systemd.services.postgresql.postStart =
+      with config.services.postgresql;
+      pkgs.lib.mkBefore ''
+        PSQL="psql --port=${toString port}"
+        while ! $PSQL -d postgres -c "" 2> /dev/null; do
+            if ! kill -0 "$MAINPID"; then exit 1; fi
+            sleep 0.1
+        done
+        if test -e "${dataDir}/.first_startup"; then
+          createdb example
+          pgbench example -i -s 50 --foreign-keys
+        fi
+      '';
 
     networking.firewall.allowedTCPPorts = [ 5432 ];
   };
diff --git a/postgrest/shell.nix b/postgrest/shell.nix
index b05d69d..bdbfa0d 100644
--- a/postgrest/shell.nix
+++ b/postgrest/shell.nix
@@ -36,6 +36,41 @@ let
 
         nixops ssh -d pgrbench client k6 run --summary-export=$filename.json - < $1
       '';
+  clientPgBench =
+    pkgs.writeShellScriptBin "pgrbench-pgbench"
+      ''
+        set -euo pipefail
+
+        # uses the full cores of the instance and prepared statements
+        nixops ssh -d pgrbench client pgbench example -h pg -U postgres -j 8 -T 30 -M prepared $*
+      '';
+  repeat =
+    pkgs.writeShellScriptBin "repeat"
+      ''
+        set -euo pipefail
+
+        number=$1
+        shift
+
+        for i in `seq $number`; do
+          echo -e "\nRun $i:\n"
+          $@
+        done
+      '';
+  pgBenchAllPgInstances =
+    pkgs.writeShellScriptBin "pgrbench-all-pg-instances"
+      ''
+        set -euo pipefail
+
+        for instance in 'm5a.large' 'm5a.xlarge' 'm5a.2xlarge' 'm5a.4xlarge' 'm5a.8xlarge'; do
+          export PGRBENCH_PG_INSTANCE_TYPE="$instance"
+
+          pgrbench-deploy
+
+          echo -e "\nRunning on a $PGRBENCH_PG_INSTANCE_TYPE\n"
+          $@
+        done
+      '';
   ssh =
     pkgs.writeShellScriptBin "pgrbench-ssh"
       ''
@@ -63,6 +98,9 @@ pkgs.mkShell {
     k6
     ssh
     destroy
+    clientPgBench
+    repeat
+    pgBenchAllPgInstances
   ];
   shellHook = ''
     export NIX_PATH="nixpkgs=${nixpkgs}:."
@@ -71,5 +109,8 @@ pkgs.mkShell {
     export PGRBENCH_WITH_NGINX="true"
     export PGRBENCH_WITH_UNIX_SOCKET="true"
     export PGRBENCH_SEPARATE_PG="true"
+
+    export PGRBENCH_PG_INSTANCE_TYPE="t3a.nano"
+    export PGRBENCH_PGRST_INSTANCE_TYPE="t3a.nano"
   '';
 }