Christian Völker
2014-Dec-07 17:25 UTC
[Gluster-users] GlusterFS Cluster over Slow Lines?
Hi all, I have not found an answer to my questions regarding GlusterFS so I decided to ask here. We have two sites which are connected through a slow 10Mbit line. But I want users accessing shared files on both sites with optimized performance. My first thought was to use drbd for this but then I would need a clustered file system(which would additionally put load on the connection) and I would have slow write access on both sites. So I am currently testing GlusterFS- am I right about the following? -Reads will always be send from the local disk -Writes go through local disk first and are replicated then. When does the client get confirmation of succesful write? Once the data is fully replicated or earlier? In first case it would slow down writes as the nodes are connected only through a small 10Mbit line, correct? -How does glusterFS handle a dual-head situation? In case the connection between the two nodes goes down? what happens when connection is back then? -I want to have access to my data on both sites (nodes), so for me a replicated volume would be fine? -thhe term "distributed" means the data is just spread across the local bricks, right? So having a single brick protected by hardware raid, is fine, too? I am not sure about how many bricks are recommended to use for what scenario. Well, loads of questiosn, thanks for your patience.... Greetings Christian
Hey, I have a similar issue, over a slow line glusterfs seemed to divide the speed by two for writes. So perhaps you might look into geo-replication ? I haven't found a solution yet. *following this topic* Svenn 2014-12-07 18:25 GMT+01:00 Christian V?lker <chrischan at knebb.de>:> Hi all, > > I have not found an answer to my questions regarding GlusterFS so I > decided to ask here. > > We have two sites which are connected through a slow 10Mbit line. But I > want users accessing shared files on both sites with optimized performance. > > My first thought was to use drbd for this but then I would need a > clustered file system(which would additionally put load on the > connection) and I would have slow write access on both sites. > > So I am currently testing GlusterFS- am I right about the following? > > -Reads will always be send from the local disk > -Writes go through local disk first and are replicated then. When does > the client get confirmation of succesful write? Once the data is fully > replicated or earlier? In first case it would slow down writes as the > nodes are connected only through a small 10Mbit line, correct? > -How does glusterFS handle a dual-head situation? In case the connection > between the two nodes goes down? what happens when connection is back then? > -I want to have access to my data on both sites (nodes), so for me a > replicated volume would be fine? > -thhe term "distributed" means the data is just spread across the local > bricks, right? So having a single brick protected by hardware raid, is > fine, too? > > I am not sure about how many bricks are recommended to use for what > scenario. > > Well, loads of questiosn, thanks for your patience.... > > Greetings > > Christian > > > _______________________________________________ > Gluster-users mailing list > Gluster-users at gluster.org > http://supercolony.gluster.org/mailman/listinfo/gluster-users >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20141208/e52e5394/attachment.html>