On 10/04/2011 20:17, Meisam Mohammadkhani wrote:
Dear Gordan, Actually our little tests showed that replicated files, could be accessible in more than one minute with GlusterFS.
I suspect the reason you saw that delay is because you put the files on one node, and it needed healing when you accessed it from the other node. If both your nodes are in sync, the access time should be reasonably transparent - certainly not a matter of minutes. Either you had a configuration error of some sort (you haven't posted your GLFS volume configs and the details of your server/network setup), or the files you were trying to access were replicating themselves to the machine you were accessing them from.
It's a big delay for us, but maybe our tests was not configured nicely. Our application is a .net application and we used CIFS to accessing the GlusterFS files.
While it would be reasonable to expect a considerable performance penalty in this kind of a setup, the latencies still shouldn't be in the minute range unless the data was replicating when you were accessing it. Have you made sure the data is synced between the nodes? Also, is your Samba using CTDB or vanilla TDB?
Also machines hardware are not so powerful. I wanna to know what should be the accessibility delay range in different cases? Can GlusterFS support "High Availability" in a fastest way as it's possible?
I think you need to outline what your expectations are. GlusterFS can do quite well in terms of bandwidth, but latency is going to be visibly increased, and you will get additional access latencies from working with CIFS shares. But as I said, this could be seconds but certainly not minutes unless your hardware is a decade out of date.
You'll have to post more details about your server topology before a more informative suggestion can be provided. But in any case, you should expect to see latencies orders of magnitude greater than with local file system access.
Gordan
