September 18, 2006 at 3:23 pm
Hi there
Anyone here who have experience with other clustering tools than MS cluster services. Had a presentation of Polyserve the other day, and that looked like a very cool product but quite expensive as well.
Best regards
SUN
September 19, 2006 at 9:03 am
Not technically a cluster product, but filling a niche just below clustering and above hand mirroring, there's XOsoft (now CA XOsoft, and I hope the CA aquisition doesn't send what has been a dream product down the wrong road.
It's remarkably easy to set up, and the advantage it has over things like Neverfail and Polyserve is they do not assume that all access is from Windows clients and that you control the methods used to connect.
I have a server in one city, an identically configured one in another. XOsoft is installed on both. A "scenario" is created that I have set up to keep the remote server's disks up to date for all databases I do not exclude (the only one I exclude is tempdb, since that gets remade at server boot anyway - and an ISV product we use is nasty on tempdb). Obviously you need bandwidth between the sites.
I also have it keep copies of my disk-based dumps.
One server is the master. The the XOsoft software understands how to start and stop SQL Servers, how to write disk-block level changes UNDERSTANDING the need for write order preservation and for transactional atomicity.
Switchover can be automatic or manual. Press the "switchover" button, it checks the state of the scenario, stops the scenario, shuts down SQL Server on the primary, brings up SQL Server on the secondary, and does a dynamic DNS update so that the network knows the secondary by the name of the first.
This is kind of a Grail for me, for 20 years I've wanted something that could do this, and homebrew attempts and other attempts have always ended up causing more trouble than they solve. This one basically just works.
There's also a module called "assured recovery" which lets you bring up the secondary for testing, and then bring it down and put it back to its pre-brought-up state, so the queues that have backed up can drain into it and a full sync is not needed.
If one side detects something on the other side that makes it questionable what the status is, it starts a sync.
In a sync, instead of just passing along changed blocks, all blocks are compared. But ONLY changed blocks need to go accross the network. It takes me several hours to do a full sync of about 150 GB.
The benefit of this over mirroring is how much it does for you. I don't configure this on the database level. The replication INCLUDES the master database. If I create a new db on the master, change permissions, move something, this all gets done for me on the backup.
I'm not sure of price - I think about $10K per server pair - a lot less than real clustering. The downside is that you do lose connectivity during the switchover, and if your apps (especially service apps!) don't make ongoing retries after losing a connection, they lose. Also, web servers like Apache tend to resolve the IP of the dataserver themselves and then CACHE THE IP! So they don't switch over. I consider that a bug, they consider it a security feature (you needn't worry about someone spoofing the server name...)
It works well for my heterogenous world...
roger reid
Roger L Reid
September 19, 2006 at 12:09 pm
Hi Roger
The product you mention is that only 2 party system?, it sound like a cool form of replication. What are the hardware requirements, do you need identical hardware like in MS clustering?
The benefit I see in polyserve is that it can handle many physical servers being part of a cluster group, can this product you mention do the same?
Thanks for your reply
Best regards
SUN
Viewing 3 posts - 1 through 2 (of 2 total)
You must be logged in to reply to this topic. Login to reply