HP developers move some features from Compaq’s version of Unix to its own HP-UX, an important part of the company’s plan to eventually merge the two product lines. This was known to happen since Day 1 of the merger of HP-Compaq. Additionally, read our recent interview with an HP-UX engineer.
HP-UX Gets Key Features from Tru64
2003-07-24 Unix 10 Comments
Thuis can hardly be a Bad Thing(TM) unless they screw up any APIs and can’t see that they’d do something like that. Most of the stuff getting merged seems to be “under the hood” anyway.
Seems bizarre. HP already offers a clustering solution, called Serviceguard http://h18000.www1.hp.com/solutions/enterprise/highavailability/lin… and has even ported it from HPUX to Linux, for goodness sakes.
And the part about a new filesystem is also bizarre, since HP has stated that they are moving to Veritas VXVM along with the long standing Veritas VXFS. These two products have been ported from the PA-RISC platform and run with the new versions of HPUX 11i 1.6 for Itanium.
I wish there was more meat on this one. I am curious on the details
Well.. here is the thing. Serviceguard is not that great when it comes to scaling or advanced features. From what I’ve read, it sounds like they will offer both products,and perhaps retire Servicegard down the road. TruCluster is better anyways, it has a global filesystem, global devices, shared memory transport(good for Oracle RAC), and lots of other features you won’t find from Sun, IBM, Veritas, and non-Ex-Dec products. It’s been around for a while, and is very mature.. Oracle based RAC off of TruCluster’s technology. Advfs was developed by DEC and Veritas back in the day. It has a bunch of features that VxFS still doesn’t have, even though they have similiar roots.
Honestly, I’m surprised that HP was able to port much of what’s under the hood, since the HPUX and Tru64 kernels are very different. I’m sure there was a lot of rewriting, especially when it came to the kernel level api’s and what not. The real test will be how HPUX customers react to the product when it’s released.I’m sure it’ll change how things are done. The entire Itanium thing seems uncertain anyways, considering how few systems have sold. I don’t think people want to switch archs in this economy.
i feel another sick “SCO allegation/litigation/intimidation” headache coming on!
>Well.. here is the thing. Serviceguard is not that great when
>it comes to scaling or advanced features. From what I’ve
>read, it sounds like they will offer both products,and
>perhaps retire Servicegard down the road.
Ok… we have a little mis-informed opinion here I think.
First of all HP’s MC/ServieGuard is currently the BEST HA Clustering solution on the market. BY FAR… Believe me I work as a UNIX Admin in a Fortune 50 and we use Sun Cluster, Veritas, and MC/SG…and MSCS on NT..And MC/SG blows them all away for reliability (ROCK SOLID), capabilities (32 nodes), and ease of use.. its very, very logical to admin. Many people have no idea about MC/SG because its limited only to HPUX (#1 in UNIX market btw, not Solaris) until recently when it was ported to Linux.
The second thing is you don’t realize that your talking about 2 TOTALLY different products:
1) HP MC/ServiceGuard is a HA Cluster meaning that services FAILOVER from one system to the other providing failure protection. This is used for things like databases, web servers etc..anything really that runs on a regular system can be made Highly Available using MC/SG.
2) HP TruCluter is more like a linux “beowolf” cluster in that you use it to make a system appear more like a single system image and run parallel-computing applications. This software instead of doing “HA type Failover” makes several boxes look like one to the application. This is really just like IBM has been doing for years with their SP Clusters on AIX.
The 2 are totally different types of clustering technologies and are typically used for very different applications. For many things the overhead involved in doing multi-system single-image clustering is to high a price for application performance, when a HA Cluster would work better. In some applications single system images are nicer, since they can take an application that could benifit from parallel computing which is not specifically written for parallel computing and spread its load out over several systems.
Thanks for your reply, it gave me alot better view of the capabilities of both TruCluster and MC/ServiceGuard.
Going by your post it would seem that TruCluster is in many ways similar to OpenMosix
This is incorrect trucluster was designed for HA. It was hacked on to the clusters to help manage the systems. It in no way helps parrallel computing. Besides providing some _manageablity_.
Basically in Trucluster the OS knows a good deal about the failover and the cluster nature of the system. Service gard is more like HACMP or the linux ones. A bunch of scripts and the application handles the failover/locking etc.
Trucluster the os does all the work and moves it for you. It is more rebust then a bunch of scripts but it is complex. It supports sharing root for management at the same time as increasing the availablity. It is more like what you would get if you integrated a failover cluster instead of glueing it together with tape. Not clear if that increases the level of availablity or not.
It basically borrows from VMS clustering ideas.
ServiceGuard used to be the best HA solution on any Unix until TruCluster was released a few years ago. TruCluster is the Unix Sysadmin’s dreams come true. With TruCluster, you don’t need to update all the scripts controlling the cluster on all the servers part of the cluster, because they share the same system disk. Administration of the servers and cluster is made a lot more easy with this feature. Also, if your application supports it all the servers in the cluster can be seen as one server. All the communications to that single IP address are redirected, in a round-robin manner by default, to each server in the cluster. If your app doesn’t support that, then you can configure the cluster to behave as a normal Unix cluster. TruCluster gives you a lot more flexibility and availability. Pretty much like VMS clustering, i.e. the only real clustering solution before TruCluster.
AdvFS is a lot more flexible than VxFS. With VxFS, you have a one to one relationship between the volume or the partition and the filesystem. With AdvFS, you can have as many volumes or partitions in a single filesystem, which means that if you want to grow your fs, you only need to add a new volume to it, rather than trying to grow one. From my point of view, it is always safe to add a new volume than to touch existing ones. Also if you have some spare space on a different diskgroup, it doesn’t care, you can add that space to the filesystem as it will be presented as a volume.
I have worked with ServiceGuard/VxFS for 5 years and loved it until TruCluster came along. Technically, if HP can pull this off, the new HP-UX is going to rock. Now wether they will be able to take advantage of that or not is another matter……
Sorry but Solaris is more better!
HP admins have bigger man parts than solaris admins