Fog storage node selection by latency instead of load
-
Hello!
I have put a fog manager with two storage nodes in 3 different locations, so they can sync automatically the images between them.They all are accessible to each other but the local network is faster and the nodes have no load issues, at most one client at a time.
I have changed the tftpboot default file to serve the IP of the node in the local network, so the machines boot fast from them, as if they where a standalone FOG. After boot, they select the storage nodes at random.
Can I “fix” the IP of storage node like I did with the boot? Or it could be done by changing the load balancing logic from the default to a latency-based one? Like ping the client and select the fastest one when there is a tie on the other values.
-
@eliphas Install the location plugin into the master node. Then associate the storage node with a location and then when you register the target computer define a location for them. Then your clients will reach out to the defined storage node. Without the location plugin the target computes will pick at random all storage nodes in the master nodes storage group. There is no real intelligence when selecting storage nodes in the storage group.
-
Hi George thanks for replying!
In this scenario I am using FOG as a base-image-deployer library for VMs, seldom registering a machine on FOG; basically deploying multiple instances of same/different images over and over, after we made the golden image.
I have read that the location plugin only works for hosts that are registered, right? -
@eliphas said in Fog storage node selection by latency instead of load:
I am using FOG as a base-image-deployer library for VMs, seldom registering a machine on FOG
Ok so you are using fog in a system builder mode. Where you just pxe boot and then select deploy image from the FOG iPXE menu (i.e. load and go). In this case the location plugin will not help you.
Are the master node and storage node at the same location or different geographic locations? If they are different locations and you are using the load and go method, you may be better served by installing a full fog server at each location, then on the HQ location add the two remote FOG servers as storage nodes. This way the images will replicate from the master HQ fog server to the remote FOG servers. The only thing you will need to do is export the image definition from the master node and then import it into the remote fog servers.
-
@george1421 said in Fog storage node selection by latency instead of load:
Are the master node and storage node at the same location or different geographic locations? If they are different locations and you are using the load and go method, you may be better served by installing a full fog server at each location, then on the HQ location add the two remote FOG servers as storage nodes. This way the images will replicate from the master HQ fog server to the remote FOG servers. The only thing you will need to do is export the image definition from the master node and then import it into the remote fog servers.
Hello George, and thanks again for the help!
Yes, they are in 3 different regions. I added the local storage nodes so they could get the images faster and still would have only one centralized place to upload the images, but did not work out as expected.
Just reinstalled the two slaves today, now they are full servers. Was thinking, if I add the slave1/slave2 as storage nodes to the master1 again, would not the master1 still point the clients to them?
Regards,
Eliphas -
@eliphas said in Fog storage node selection by latency instead of load:
Just reinstalled the two slaves today, now they are full servers. Was thinking, if I add the slave1/slave2 as storage nodes to the master1 again, would not the master1 still point the clients to them?
In the above configuration (since they are all now full fog servers) the image raw files will replicate from the master to the storage nodes (full remote fog servers). What will not happen automatically is the image definitions in the database will not be auto synchronized. You will need to do an export using the web gui on the master node and then an import on the storage node to get the image definitions.
Ok now to the clients. Since there is a full fog server at each site, you will have each site configured to talk to its local FOG server. The central server is not in the picture when pxe booting or imaging the remote sites.
