liufuyang commented 5 years ago

Simple comparison between TiKV and Bigtable on GCP

TiKV is in k8s cluster, 4 nodes, each is a standard v4 node. Bigtable is a standard 3 node cluster.

Result: Bigtable left, Tikv right

(read and write count is print out every seconds) Seems Bigtable is faster on write, TiKV is faster on read, in this test setup.

Test client is this java main file running

for bigtable test, it is running on an empty node (with code version master of this repo)

for tikv test, it is running on the node pd is on (with code version branch java-gcloud of this repo)

kubectl get pods  -o wide -n tidb
NAME                              READY   STATUS    RESTARTS   AGE     IP           NODE                            NOMINATED NODE
demo-discovery-84c889bf96-5x7lv   1/1     Running   0          4h48m   10.40.5.4    gke-tidb-pool-1-60d488aa-kk1v   <none>
demo-monitor-5ddb4bf8c-khbmg      2/2     Running   0          4h48m   10.40.4.3    gke-tidb-pool-1-60d488aa-5k8n   <none>
demo-pd-0                         1/1     Running   0          4h48m   10.40.7.5    gke-tidb-pool-1-60d488aa-v256   <none>
demo-tikv-0                       1/1     Running   0          4h48m   10.40.8.3    gke-tidb-pool-1-60d488aa-8t1d   <none>
demo-tikv-1                       1/1     Running   0          4h48m   10.40.4.4    gke-tidb-pool-1-60d488aa-5k8n   <none>
demo-tikv-2                       1/1     Running   0          4h48m   10.40.6.5    gke-tidb-pool-1-60d488aa-t2dd   <none>
kvrawclienttest                   1/1     Running   0          31m     10.40.5.27   gke-tidb-pool-1-60d488aa-kk1v   <none>
kvrawclienttest-tikv              1/1     Running   0          23m     10.40.7.6    gke-tidb-pool-1-60d488aa-v256   <none>

Other info:

CPU load on test node

TiKV load while testing (above)

Bigtable load while testing (above)

c4pt0r commented 5 years ago

Awesome job! Does TiKV run on persistent disk? or local SSD?

liufuyang commented 5 years ago

Round 3 - key, value corrected

4 node write to Bigtable, then all read from Bigtable
4 node write to TiKV, then all read from TiKV

Bigtable write

11:52:16,474 [main]  INFO MainBT:130 - [Bigtable Write]   4058 total actions, avg time = 15801 μs
11:52:17,475 [main]  INFO MainBT:130 - [Bigtable Write]   4011 total actions, avg time = 15922 μs
11:52:18,477 [main]  INFO MainBT:130 - [Bigtable Write]   4004 total actions, avg time = 15937 μs
11:52:19,478 [main]  INFO MainBT:130 - [Bigtable Write]   4020 total actions, avg time = 15951 μs
11:52:20,479 [main]  INFO MainBT:130 - [Bigtable Write]   3426 total actions, avg time = 18385 μs
11:52:21,480 [main]  INFO MainBT:130 - [Bigtable Write]   2710 total actions, avg time = 23395 μs
11:52:22,481 [main]  INFO MainBT:130 - [Bigtable Write]   3778 total actions, avg time = 16974 μs
11:52:23,482 [main]  INFO MainBT:130 - [Bigtable Write]   3830 total actions, avg time = 16684 μs
11:52:24,483 [main]  INFO MainBT:130 - [Bigtable Write]   3823 total actions, avg time = 16770 μs

TiKV write

11:53:15,629 [main]  INFO MainTiKV:112 - [TiKV Write]   3966 total actions, avg time = 16157 μs
11:53:16,630 [main]  INFO MainTiKV:112 - [TiKV Write]   4530 total actions, avg time = 14015 μs
11:53:17,632 [main]  INFO MainTiKV:112 - [TiKV Write]   4782 total actions, avg time = 12821 μs
11:53:18,633 [main]  INFO MainTiKV:112 - [TiKV Write]   2581 total actions, avg time = 23983 μs
11:53:19,635 [main]  INFO MainTiKV:112 - [TiKV Write]   4630 total actions, avg time = 13757 μs
11:53:20,636 [main]  INFO MainTiKV:112 - [TiKV Write]   3152 total actions, avg time = 20406 μs
11:53:21,637 [main]  INFO MainTiKV:112 - [TiKV Write]   3201 total actions, avg time = 19963 μs
11:53:22,638 [main]  INFO MainTiKV:112 - [TiKV Write]   4349 total actions, avg time = 14621 μs
11:53:23,640 [main]  INFO MainTiKV:112 - [TiKV Write]   3281 total actions, avg time = 19634 μs
11:53:24,641 [main]  INFO MainTiKV:112 - [TiKV Write]   3886 total actions, avg time = 16189 μs
11:53:25,642 [main]  INFO MainTiKV:112 - [TiKV Write]   4159 total actions, avg time = 15751 μs