We're currently investigating converting some of our HP-UX systems to Linux. The primary driver is cost of hardware and support. I know this is a fairly open-ended question, but has anyone done this and could they share their experiences, good or bad? We run Oracle and an ERP for the most part.
You really don't convert systems from HP-UX to Linux. You build new systems and migrate data and functionality. But I knew what you meant.
I've been in a number of situations like this and in some cases the "conversion" was called off.
Yes, HP-UX hardware costs more. However HP-UX hardware does more than typical Wintel hardware. An Itanium based system will run circles around a similarly configured Wintel Linux system. You get more bang for your buck.
You will probably need MORE Wintel CPU's on Linux to do the same amount of work the HP-UX oracle server. Oracle's pricing model is $40,000/CPU and it does not take into account CPU strength.
So where this works best for cost savings us under utilized systems. But a possible good alternative is HPVM. It will let you run HP-UX, Linux and Windows on the same hardware and allocate CPU in chunks as small as 1% of a CPU. A careful consolidation can save you a lot in licensing fees.
I would urge a good analysis be done. How much money will be saved, how much will have to be spent to get equivalent levels of service and license fees.
SEP
Steven E Protter Owner of ISN Corporation http://isnamerica.com http://hpuxconsulting.com Sponsor: http://hpux.ws Twitter: http://twitter.com/hpuxlinux Founder http://newdatacloud.com
Yes, I realize there's not a literal software conversion between the two, I meant that more from a logical or strategic level. My post should have been written more clearly.
Steven, I'm glad you replied as I was hoping to get in contact with you on this topic. My company was looking to engage someone about this. I tried contacting you through your ISN web site.
No email yet. Probably that presumptive trigger happy on steroids spam filter.
steve at hpux.ws
Change the at and spaces to a @
Steven
Steven E Protter Owner of ISN Corporation http://isnamerica.com http://hpuxconsulting.com Sponsor: http://hpux.ws Twitter: http://twitter.com/hpuxlinux Founder http://newdatacloud.com
We are in the progress of doing just what you describe. We have moved a number of servers from HP-UX to SuSE Linux (SLES) and so far we have been pleased with the results.
The justifications we used are the same that you mentioned.
The types of servers we have moved to SLES have been some infrastructure things like e-mail, ftp, DNS.
We have also migrated a couple of applications servers including our IBM Tivoli Workload Scheduler server (formerly Maestro) and our EDI server which has an Oracle DB back end.
In both cases we took the opportunity to upgrade the applications and DBs to the latest versions.
In the case of the Maestro server, all job schedules, etc. moved over without a hitch.
In the case of the EDI server there were some issues with some shell scripts and differences in the Bash shell (linux) versus the Posix shell (HP-UX) which required some minor tweaking on Linux. Not much Oracle data was migrated in our case.
When licensing Oracle, yes the base price for Oracle Enterprise Edition is $40,000 per CPU. However, there are certain "multipliers" for multi-core processors.
I **THINK** that Wintel systems are 0.5 per core and Itanium/PA-RISC are 0.75 per core.
So, with a Wintel server, if you have 2 x quad core processors (8 cores), then you would pay licensing equivalent to 4 cores (8 x 0.5). That can make Oracle licensing slightly less frightening, but you would need to verify this with Oracle.
We are planning, I think, to eventually migrate all HP-UX to Linux.
Over the past few years, I've migrated 4 of our 5 production HP-UX systems to Red Hat Linux. Working on migrating the last one now as I write. So far, been able to work each migration into a standard suite of bash shell scripts (supporting a multi-DB environment). While the primary catalyst for migrating was cost-savings, performance also went up all around... mostly due to standard hardware technology gains (CPU, RAM, disk).
Our migrations - which were also for ERP (non-Oracle) - are from HP 9000 HP-UX 11.x to Red Hat Enterprise Linux on Intel hardware, previously to standalone HP DL 580 servers, but now to virtualized (VMware) clients.
For reference, we have fairly modest user counts (low 100s) and DB sizes (low 20s GB).
Dirk makes an apples to orange performance comparison.
If you take a brand new Wintel box and use it to replace a 5 year old PA-RISC box, you will get better performance.
Take an IA-64 up to date system CPU for CPU, it will and does in our tests here run circles around a brand new Wintel box.
Itaninium is a much maligned, but extremely well designed, powerful architecture.
SEP
Steven E Protter Owner of ISN Corporation http://isnamerica.com http://hpuxconsulting.com Sponsor: http://hpux.ws Twitter: http://twitter.com/hpuxlinux Founder http://newdatacloud.com
>>Take an IA-64 up to date system CPU for CPU, it will and does in our tests here run circles around a brand new Wintel box.
Yeah, but you're going to pay 5x (at least) the $$$$ for the Itanium (HW & SW) as you would for the Wintel/Linux combination.
If one is looking at Oracle you can get a 3 server Oracle RAC cluster for the same or fewer $$$$ as a single Itanium server with HP-UX. At least this was true for the price comparisons I did last summer.
As a SAP shop, we are slowly moving everything non-clustered to SLES on ProLiant. The price/performance is just a lot better compared to our current IA64/Montecito servers.
For misson critical systems we'll keep the central components (SAP CI and Oracle) on HP-UX ServiceGuard Metroclusters and if needed use SLES systems as the application servers.
With Linux you will probably miss HPUX tools like ioscan, Ignite/UX and the excellent documentation... TANSTAAFL ;-)
At our customer we are migrating everything from HP-UX to Linux. (to mostly SLES 9s and 10s) Apart from some exotic HA SG packages everything can be migrated with more or less problems. Usually the user scripts must be modified to get them to work with the bash shell which is the default in linux.
This is a large heterogenous server environment with ~2000 linuxes and the HP-UX population was degraded from 400 to 50 machines during the last 3-4 years. There are also some AIXes left, Solaris wears relative well with around 200 virtual systems.
@steven: sure but it depends on what you are comparing. i recon price is the main comparison in most cases, so compare a equal priced x86-64 box against IA64 and then you have a total different outcome. even if comparing between cpu's then the question remains is the IA64 that times more performant that it justifies the price? how long will it keep this lead? x86-64 is developing faster and is a bigger market. on top of that, linux on x86-64 is the ideal platform for horizontal scaling, which is the way things scale these days, even db's.
I helped a customer last year migrate their core broking system from HP-UX to Linux. The migration isn't quite complete yet, but they've done several countries so far.
There were a few small issues, mostly to do with custom shell scripts for administration and menus. They had a legacy of "everything will work with ksh only", so occasionally there were things that bash choked on.
We bit off a lot more than we needed to. For example, we chose also to move authentication out of /etc/passwd and to use ActiveDirectory instead at the same time. Also I insisted on making everything build automatically without intervention so that every environment would be consistent, which is not something they had done much before. These two requirements caused more problems than the script issues.
(But on the other hand, it meant that I could redo the build dozens of times on a virtual machine on my desktop until we had it sorted. This would have been possible on an Itanium virtual machine too, but it would have been harder to price justify buying a machine big enough to let us have a crash-and-burn play area.)
Other bad things: we found we couldn't trade in licenses from HP-UX to Linux for that particular application, so that bumped the cost up a little.
Price-wise it was still a no-brainer though. The systems that were being replaced were nowhere near fully utilised. They were only being replaced because they were out of support. There was no way of justifying the price of an Itanium box when a ProLiant worked equally well. The Proliants were much, much faster than the PA-RISC boxes they replaced.
To make it more interesting, these boxes were in sites where HP enterprise engineers were essentially unavailable. So hardware support became much, much easier because even in the middle of nowhere it is possible to get hardware engineers that can replace a disk in a ProLiant.
We went overboard with memory because we wanted as much as possible cached in RAM. I think we worried about the ProLiant boxes having worse I/O performance than the HP-UX boxes, but that turned out not to be a problem at all.
On a related note, back in 2007 one of my customers wanted a course for their intrastructure architects who knew how to architect big HP-UX solutions but wanted some suggestions of how to architect Linux-based solutions. It's a bit out-of-date (and I don't think it was one my best courses anyway), but the slides from it are at http://www.ifost.org.au/Documents/architecting.pdf
