Slashdot Mirror
Domain: polyserve.com
Stories and comments across the archive that link to polyserve.com.
Comments · 12
-
Re:Free is not necessarily as in free beerIf you want commercial software, HP will sell you quality-assured Lustre and decent hardware in HA configurations, relabelled "HP SFS". Also, in the spirit of full-disclosure, I am a supporting engineer of Polyserve, now owned by HP.
If you're looking for a HA storage solution, have you looked at Cluster Gateway? It's essentially a Polyserve file system with the NFS or CIFS solution pack, depending on which platform you're implementing. The software-only costs are relatively low (I've been bitching for a while that they're giving it away,) and you can use commodity servers and storage.
A scalable, clustered file system, that if properly implemented (the important part,) is single-point-of-failure immune. The minimum is two nodes - that's a scale-down if I ever heard one. It works with iSCSI (Openfiler is used by me in my test labs) or FC storage.
Think of it as GFS + Redhat Cluster Suite, but better implemented. On the other hand, if you're looking for zero-dollar-investment, check out Cluster Suite and GFS with CentOS for free. The user interface is terrible, and simple tasks are made hard, but it does work well, again, if implemented properly. -
Re:Yes, Oracle will get into the OS biz. Here's wh
But wasn't TruCluster a reimplimentation of the VAXcluster layered product?
Yep. Though its not quite as good as the cluster filesystem on OpenVMS. Any cluster member can read files (of 64k or larger) direct from shared storage, but asynchronous writes are still funneled over the cluster interconnect to an AdvFS domain server. So its important for optimum performance (and to avoid flooding the Memory Channel interconnect) to align the server for a particular AdvFS domain with the cluster member doing the most I/O to that domain. However, AdvFS on TruCluster does support Direct I/O, so if an application has multiple instances running on different cluster members, and they can communicate with each other to coordinate I/O as Oracle RAC does over the cluster interconnect (when its not using ASM) then the app can open files with using O_DIRECTIO and write to them directly, bypassing the AdvFS domain server and AdvFS buffer cache.
The only other Unix Cluster Filesystem I've seen that I would compare to OpenVMS and doesn't have this asynchronous limitation is the Symetrical Cluster File System from Polyserve. Though I'm a bit wary of proposing them as a solution to Linux customers as their products also run on MS Windows, and I'd be very surprised if MS doesn't gobble them up at some point in the future and dump the Linux port.I, as a long-time OpenVMS user, can't imagine not having a cluster-aware FS.
Ah, but you, like me, have "grown up" with an OS that implements SSI (Single System Image) natively. We know the benefits of having cluster common system files and application directories. For example a common account/password database without having to resort to NFS or similar. But the rest of the industry have never experienced this and frankly don't know what they're missing. As you may be aware, HP were planning to port TruCluster functionality from Tru64 over to HP-UX, but last year it got canned. My understanding of events is that this decision was customer driven. I was told that HP held workshops with their top customers and asked them what it was they wanted. The message they got back from that was their customers didn't like the fact HP were about to change HP-UX into something almost unrecognizable that would force the majority of their customer base to re-learn what they already knew. They wanted to move forward with a "unix standard" CFS implementation (i.e. Veritas) and didn't see SSI as a big enough reason to make such a huge change. Veritas had a product almost ready that meant HP-UX could get to market with a CFS a year ahead of when the TruCluster port was due to finish, so the decision by HP to drop it was a no brainer. The majority of their customers just didn't want it! Once TruCluster finally dies (it's not quite yet, you can still buy it) I doubt you will ever see another another commercial Unix SSI cluster solution. The industry just doesn't seem to see the need.How do database apps running on RAC nodes share scripts, flat files, etc without a DLM and cluster-aware FS?
Well, I'm no Oracle expert, but as I understand it if you're running Oracle ASM, then from inside Oracle you can change directory, and view files (cd, ls) from any node in the cluster as if you were running an OS native cluster filesystem. So once oracle is up and running it can look to itself for any node specific scripts it needs. Plus it has its own DLM.
As for app binaries, Oracle recommends you have an 18GB filesystem (disk or partition) that is private to each RAC cluster member and the Oracle binaries and startup files are located in there. The Oracle RAC installation process then copies parts of the Oracle kit that need to be cluster common into a common area which I think exists inside of ASM. 18GB is nothing these days, and you can easily create this on a SAN and make it private to a system using selective storage presentation and SAN Zoning from the SAN of your choice. So having multiple copies of some Oracle files on your cluster is not considered a problem. -
Re:Whither OCFS2 ?
GFS is not the only CFS in town for Oracle RAC. Have you heard of MatrixServer? They've had an ODM api and true CFS for RAC for years now. They've also have gone through the new Linux RAC certification test, something GFS cannot claim.
http://www.polyserve.com/pdf/10gRAC_MxS_Whitepaper .pdf -
Re:downtime during backup?
OpenVMS was around in 1977? I mean, it is great that VMS had a rich feature set, but how does that help those of us that manage enterprise environments today? Since the original context of this thread was backup on the Windows server platform, how about something really useful. Check out an OpenVMS based product from PolyServe that actually works on a current day, pertinent OS's with real market share.
-
Re:Solaris Vs Linux?
Sorry sir, but your post was a crock of crap. Had to be said.
There's this thing called "fork and exec" which has been out for awhile, which very easily enables an application to scale to N CPUs. Apache for example, will nicely scale to lots of CPUs assuming the underlying OS efficiently does copy-on-write, thread/process management, etc. Solaris does.
If you believe "Oracle didn't scale on Sun E10Ks period", check out the site called eBay. It's the only way they are able to handle the massive workload...
Oracle is pushing clustering now for the reason a previous poster gave- Cheaper hardware means more $$ for licensing, with a static budget.
Lastly your claim about Oracle scaling effectively across 120-240 Linux CPUs appals me. Are you claiming that RAC can be deployed to 30-60 quad-CPU boxes? 15-30 8-CPU boxes? You may be interested to know that 9i RAC degrades in performance beyond 3 nodes- a 3 node cluster performs better than a 4 node cluster. Oracle themselves tout RAC more as an "accessibility" technology that removes single points of failure, rather than a scalability approach. Heck, there are even companies that sell third-party tools to make RAC more scalable...
In conclusion, I do not believe you have any clue with regards to the subjects you are addressing in your post. -
Re:Solaris Vs Linux?
Sorry sir, but your post was a crock of crap. Had to be said.
There's this thing called "fork and exec" which has been out for awhile, which very easily enables an application to scale to N CPUs. Apache for example, will nicely scale to lots of CPUs assuming the underlying OS efficiently does copy-on-write, thread/process management, etc. Solaris does.
If you believe "Oracle didn't scale on Sun E10Ks period", check out the site called eBay. It's the only way they are able to handle the massive workload...
Oracle is pushing clustering now for the reason a previous poster gave- Cheaper hardware means more $$ for licensing, with a static budget.
Lastly your claim about Oracle scaling effectively across 120-240 Linux CPUs appals me. Are you claiming that RAC can be deployed to 30-60 quad-CPU boxes? 15-30 8-CPU boxes? You may be interested to know that 9i RAC degrades in performance beyond 3 nodes- a 3 node cluster performs better than a 4 node cluster. Oracle themselves tout RAC more as an "accessibility" technology that removes single points of failure, rather than a scalability approach. Heck, there are even companies that sell third-party tools to make RAC more scalable...
In conclusion, I do not believe you have any clue with regards to the subjects you are addressing in your post. -
Re:I had a related question
SAN with polyserve cluster filesystem? (sitting behind a clustered NAS if you can't do fibre to every node)
http://www.polyserve.com -
Solution
Try a cluster file system.
"Filers" create "hot spots" whereby often-accessed directories/files create IO bottlenecks.
I think you can use this CFS to create a directory tree with over 200TB of data (/home/lun1,
-
Re:Depends on what kind of a "SAN" you meanYou are quite correct. All the currently shipping products have a metadata server which introduces both a performance bottleneck and in most cases a single point of failure.
<Shameless plug>
I work for a company developing a filesystem which is truly symmetric, and hence does not suffer from these issues. If you are at all interested, check out http://www.polyserve.com/products/pmxs.html for details. It's currently in Beta.
</Shameless plug>Regards,
Tim
-
Re:BSD Clusters?
Polyserve Understudy will let you cluster any combination of FreeBSD, Linux, Solaris and NT for loadsharing and failover. You used to be able to get prices on the web site, but I can't find them there now.
-
Clustering software or management software?
If you are looking for software to create a cluster, there are several, depending upong what type of cluster you are trying to create. If you are creating a service-based cluster, check out TurboLinux Cluster Server, Linux Virtual Servers, PolyServe Understudy, and Legato. There are many others available, including hardware solutions from Cisco, F5, and Alteon. I'm not too familiar with Beowulf-type clusters.
If you are looking for software to manage groups of systems, that's a whole different story. You might look into Enlighten DSM, Tivoli, or OpenNMS. I'm sure there's a lot of competition in that field as well, but I don't have any experience with those products. -
I think we need more information here.
What kind of application are you trying to fail-over? A database? A web-server?
If you wanted a web-farm, that's dead stupid easy. Fail-over database/ftp/nfs isn't too hard, but (presently) requires commercial software. Understudy Polyserve, Wizard Watchdog, or even RSF-1 are just some of the HA clustering products available.