Skip to main content

Check connectivity of private network

  • August 4, 2015
  • 4 replies
  • 20 views

FilippoC

Hi 

 

i've installed a 3 node cluster of Vertica (community edition).

I would to know if the cluster is correctly working on private network.

The nodes (hp Blade) have two network interface, one is private network and other is public network where server application connect to it.

We have launched the installation script with the following parameters:

 

/opt/vertica/sbin/install_vertica --hosts VERT1BE01,VERT1BE02,VERT1BE03 -u dbadmin -p dbadmin --rpm /tmp/vertica/vertica-7.1.2-0.x86_64.RHEL5.rpm --dba-user dbadmin -p *********** --data-dir /opt/databases/dbadmin --dba-user-home /opt/databases/dbadmin --ssh-identity /root/.ssh/vid_rsa --control-network 100.100.7.255

 

Where:

  • 100.7.255 is the broadcast of the private interface
  • VERT1BE01,VERT1BE02,VERT1BE03 are hostname of nodes defined in /etc/hosts file with ip of production interface

 

Thanks in advance.

4 replies

Forum|alt.badge.img+2
  • Participating Frequently
  • August 5, 2015

Hi,

 

in the -s option you need to put the IP or host name for the PRIVATE INTERFACE. The public interface is the IP that you give the customers or tools to connect to Vertica. Does make sense?

 

You can validate if the network is fine by running /opt/vertica/bin/vnetperf. You can see details about this tool in teh documentation.

http://my.vertica.com/docs/7.1.x/HTML/index.htm#Authoring/InstallationGuide/scripts/vnetperf.htm?Highlight=vnetperf

 

Hope this helps,

 

Eugenia


FilippoC
  • Author
  • New Participant
  • August 6, 2015

Hi,

 

with the command:

 

[dbadmin@VERT1BE01 ~]$ /opt/vertica/bin/vnetperf

The maximum recommended rtt latency is 2 milliseconds. The ideal rtt latency is 200 microseconds or less. It is recommended that clock skew be kept to under 1 second.
test | date | node | index | rtt latency (us) | clock skew (us)
-------------------------------------------------------------------------------------------------------------------------
latency | 2015-08-06_13:59:44,447 | 192.168.21.230 | 0 | 23 | 1
latency | 2015-08-06_13:59:44,447 | 192.168.21.231 | 1 | 62 | -122
latency | 2015-08-06_13:59:44,447 | 192.168.21.232 | 2 | 87 | 508

The minimum recommended throughput is 100 MB/s. Ideal throughput is 800 MB/s or more. Note: UDP numbers may be lower, multiple network switches may reduce performance results.
date | test | rate limit (MB/s) | node | MB/s (sent) | MB/s (rec) | bytes (sent) | bytes (rec) | duration (s)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
2015-08-06_13:59:44,448 | udp-throughput | 32 | 192.168.21.230 | 30.5479 | 30.5479 | 32047104 | 32047104 | 1.00048
2015-08-06_13:59:44,448 | udp-throughput | 32 | 192.168.21.231 | 30.5438 | 30.5438 | 32047104 | 32047104 | 1.00061
2015-08-06_13:59:44,448 | udp-throughput | 32 | 192.168.21.232 | 30.5454 | 30.5454 | 32047104 | 32047104 | 1.00056
2015-08-06_13:59:44,448 | udp-throughput | 32 | average | 30.5457 | 30.5457 | 32047104 | 32047104 | 1.00055
2015-08-06_13:59:45,449 | udp-throughput | 64 | 192.168.21.230 | 61.0558 | 61.0871 | 64061440 | 64094208 | 1.00062
2015-08-06_13:59:45,449 | udp-throughput | 64 | 192.168.21.231 | 61.0588 | 61.0588 | 64061440 | 64061440 | 1.00057
2015-08-06_13:59:45,449 | udp-throughput | 64 | 192.168.21.232 | 61.065 | 61.0338 | 64061440 | 64028672 | 1.00047
2015-08-06_13:59:45,449 | udp-throughput | 64 | average | 61.0599 | 61.0599 | 64061440 | 64061440 | 1.00055
2015-08-06_13:59:46,450 | udp-throughput | 128 | 192.168.21.230 | 122.098 | 122.098 | 128057344 | 128057344 | 1.00022

 

 if i launch on private network i have this:

 

[dbadmin@VERT1BE01 ~]$ /opt/vertica/bin/vnetperf --condense -hosts 100.100.7.230,100.100.7.231,100.100.7.232

 

 

The maximum recommended rtt latency is 2 milliseconds. The ideal rtt latency is 200 microseconds or less. It is recommended that clock skew be kept to under 1 second.
test | date | node | index | rtt latency (us) | clock skew (us)
-------------------------------------------------------------------------------------------------------------------------
latency | 2015-08-06_14:06:36,282 | 100.100.7.230 | 0 | 25 | 2
latency | 2015-08-06_14:06:36,282 | 100.100.7.231 | 1 | 73 | -621
latency | 2015-08-06_14:06:36,282 | 100.100.7.232 | 2 | 81 | 691

The minimum recommended throughput is 100 MB/s. Ideal throughput is 800 MB/s or more. Note: UDP numbers may be lower, multiple network switches may reduce performance results.
date | test | rate limit (MB/s) | node | MB/s (sent) | MB/s (rec) | bytes (sent) | bytes (rec) | duration (s)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
2015-08-06_14:06:36,283 | udp-throughput | 32 | 100.100.7.230 | 30.5479 | 30.5479 | 32047104 | 32047104 | 1.00048
2015-08-06_14:06:36,283 | udp-throughput | 32 | 100.100.7.231 | 30.5458 | 30.5458 | 32047104 | 32047104 | 1.00055
2015-08-06_14:06:36,283 | udp-throughput | 32 | 100.100.7.232 | 30.5459 | 30.5459 | 32047104 | 32047104 | 1.00055

 

So how can i sure that cluster is comunicating with nodes on the private network?

thanks.

 

Filippo

 


Forum|alt.badge.img+2
  • Participating Frequently
  • August 7, 2015

Tools like iperf can do this.

https://en.wikipedia.org/wiki/Iperf


Forum|alt.badge.img+2
  • Participating Frequently
  • August 7, 2015

Hi Flippo,

 

You asked:

 

"So how can i sure that cluster is comunicating with nodes on the private network?"

 

To see which one uses vertica you need to see which one uses spread. So do

 

select * from nodes;

 

and the address that the nodes have is the one that spread uses. Check there if the IP is the private. If not you need to change it and that is something that Tech Support can help you. The Private should be in a vlan to move only the traffic from the nodes. Does this make sense?

 

Please let me know,

 

Eugenia