Oracle stop development on itanium

[Alzhy]

TProc,

Apologies if the tone and manner of my postings above seem to offend you and I am pretty sure "some". But I stand by my words - X86 systems (LINUX and S...is- hmm) are less complex and offer far reacher HA, clustering and manageability at a lower cost than their RISC based cousins. And THIS amigos is the very reason there's been (and will continue) a serious UNIX away migration over the last 3 years.

As to the overpaid Admins comment - it is true, UNIX Admins (HP-UX and AIX specially)still command the higher pay grade than their Linux counterparts and Corporates are "noticing". For UNIX admins though - it's been largely easier to make the transition to easier managed, less complex X86 Linux systems.

Hey - MEL and SEP were quoted in in a Computerworld Article today... Kudos to these 2 gents!

Here's the article:
http://www.computerworld.com/s/article/9215082/Oracle_customers_using_Itanium_chips_rethink_IT_upgrades?source=CTWNLE_nlt_entsoft_2011-03-28

[Shibin_2]

I found this link from enterprisedb blog.

http://blogs.enterprisedb.com/2011/03/28/747/

[Wim Rombauts]

Really great.

Since 90% of the software we run is home-made, we aren't married to Oracel at all. And something like EnterpriseDB is in fact what I've been looking for for years. I only didn't look at it enough in the past.

[Kenan Erdey]

Hi,

as said linux starts adressing mission critical services more, here is the link that compares linux and hp-ux with some criterias.(ok it's a paper that hp wanted to be written)

http://www.bitpipe.com/detail/RES/1297866302_813.html

[Wim Rombauts]

Alzhy,

To be honest, when we last invested in HP Integrity servers, prices were not that much higher than prices for Linux/x86.
And compared to what we yearly have to pay for software support, hardware prices are irrelevant.

[Tony Iams]

Hi all,

Thanks to everyone who participated in the poll about which alternatives might be considered for Oracle on HP-UX. I received 66 responses, which broke down as follows:

32% - Oracle on some other server platform (not Sun)
27% - IBM DB2 on HP-UX
26% - Other relational database
6% - Oracle on Sun servers
3% - Microsoft SQL Server
3% - Other database designed for unstructured data

Here is a link to the blog I posted with my analysis of the data:
http://ow.ly/4oKDh

Feel free to drop me an email if you would like to discuss this further.

Cheers,

Tony Iams
Senior Analyst, System Software Research
Ideas International, Inc.
tonyi@ideasinternational.com

[ecc_guru]

If I can't run Oracle on Itanium, then HPUX is totally useless to me. I manage a shop of multiple Superdomes running HPUX/Oracle 10g/SAP.

We have already started discussions on migrating out of HPUX to the x86 platform. I am no longer tied to HP since everybody else sells x86 anyway. As a customer I get better pricing. I really like HPUX, but more often enough it is economics that wins over technical. Integrity Superdomes & HPUX are just too expensive to maintain. Superdome equates to SuperExpensive.

HPUX is dying slowing. Good move by Oracle.

[Torsten.]

Somewhere I heard the percentage of linux and windows on Itanium was less than 5 percent, so the decision of MS and RH was reasonable.

But of course the percentage of oracle on itanium (on hp-ux) is so much higher.

I think oracle makes still good money with their products on hp-ux.

They say they stop develeopment, but isn't it stopping porting to itanium instead?

From what I heard the Integrity hardware is considerably less expensive compared to predecessors (e.g. todays Blades; Superdome 2 vs. Superdome), because they share many parts with the proliant / C-class blade family.

[Pradep]

HP need to acquire SAP or make their own DB to compete with oracleDB. No point porting HPUX to x86. If its ported, we consider other better hardware platform like IBM or even sparc. we buy HP hardware only for their high reliability. Cannot run MC apps on cheap x86.

I hate linux, whether RHEL or Oracle Linux. Its too immature/unstable OS to handle Mission critical apps. Linux is a rubbish OS suitable just for playing and learning for kids only.

Any lastly, it was evident that oracle acquired sun for Hardware and Java, not for mysql, ( contrary to the agenda in EU to stop the deal). vendors should have raised this issue that time in EU, of competition killing when HW/OS/DB/apps all coming from same vendor.

[Viktor Balogh]

> HP need to acquire SAP or make their own DB to compete with oracleDB.

I am afraid that it won't be a solution. The saying here is that the most DBAs are knowing only Oracle, and they wouldn't migrate to another database. They also cannot just change the whole DBA staff. But I'm not sure how it looks abroad, especially outside of Europe?

[James R. Ferguson]

Hi (again):

> Pradep: I hate linux, whether RHEL or Oracle Linux. Its too immature/unstable OS to handle Mission critical apps. Linux is a rubbish OS suitable just for playing and learning for kids only.

I invite you to drop the emotion and tell us *exactly* why you have concluded what you have. *THAT* would make an interesting dialog.

Regards!

...JRF...

[Torsten.]

>> *THAT* would make an interesting dialog.


I'm looking forward for this kind of posts ...

[Duncan Edmonstone]

having a bad day - 3rd time lucky:

JRF, I'll take up that challenge... I'm no Linux "expert", but if I were to list my main gripes with Linux as a platform for mission critical apps I would say:

- Lack of solid support for "online add/replace/delete" of hardware components (CPU/Memory/IO)
- HW error handling not tightly integrated to OS (here's one Sun could certainly have done a lot about with Solaris/x86, but to my knowledge didn't)
- proven significant high IO throughput (no argument with straight CPU-type performance)
- immature storage stack (doing MPIO is a headache on Linux compared to HP-UX)
- too many updates, too often
- too short support lifetimes for the OS (to be fair RedHat are getting better at this) and for the hardware
- complex support configurations when introducing third party infrastructure software (such as the Vx stack)
- Inconsistent manaement "paradigms" - hate to use that term, but it's the best I can come up with - what I mean is, every Linux admin seems to do tasks in a different way to every other
- Too many constraints on "Mainstream" kernel hardware support (e.g. last time I looked RHEL5 only supported 512GB memory/28 IO cards - no doubt it's higher now)
- Obviously Linux has excellent OS-level virtualisation capabilities thanks to VMware/KVM etc, but the in-OS virtualisation capabilities (equivalent to HP-UX SRP, or Solaris Zones) has always struck me as weak/not commonly implemented, and I'm not aware of any "workload management" tools similar to HP gWLM.
- Immature clustering technology (you should have heard the howls of protest from our customers when we discontinued the Linux version of Serviceguard and they were forced to use RedHat clustering - unfortunately it's hard to compete against something that comes free with the OS, and bean counters make too many of these decisions)

HTH

Duncan

[Kenan Erdey]

Hi duncan,

a lot things is written in the document(COMPARING HP-UX 11i V3 AND LINUX FOR HOSTING CRITICAL WORKLOADS) that i posted above somewhere. i think it's more about RAS features. unix is supported by a vendor. hw handling is done becauuse of OS knows HW. because unix more likely used in enterprise. vendors have cross compatibility between them. a lot things are ok in linux but you should work more to accomplish a task. also commercial unix is more likely to adopt new regulations. and linux is less likely to adress.IMHO :)

[James R. Ferguson]

Hi (again):

@ Duncan, as always, you have added great value to the discussion. Your comments are exactly the type of dialog we need, or at least, I wanted to hear. Thanks.

Regards!

...JRF...

[Alzhy]

Since this thread is morphing into lambasting what to me is the single biggest cause of HP-UX impending demise - I will stand in defence of the OS symbolised by the flightless Penguin ;^))

I have been amazed by the continued claim of lack of or inferior RAS as the single biggest FUD for X86/Linux platforms - well let me tell ya, I would agree if this was 3 years ago. Hello? Have you HP-UX centric admins ever actually seen or read or better yet experienced X86 systems from HP, IBM and Dell the last 3 years??

Then there's the issue of no HotPlug for CPU and Memory, "weak partitioning continuum and resource management" as well as anemic clustering and high availability -- whilst some of the RISC features like CPU hot plug are maturing, and WLM-like resource management staring to appear -- these features are not really THAT necessary as Vortualization and Private Clouds in lieue of clustering are what's making Linux increasingly doing Mission Critical tasks.

Some of the world's biggest financial markets are these days powered by Linux as a testament to its enterprise readiness.

It's unavoidable -- Linux will simply become increasingly better over the years. Coupled with advances in X86 architecture with RAS pedigree from vendors like IBM and HP -- the gap is closed between RISC/UNIX systems and Linux/X86 -- heck even Windows/X86...

And if HP Proliants and Blades still lord over the X86 market -- then HP is the de-facto Linux King.

[Duncan Edmonstone]

Alzhy,

I deliberately didn't call out "inferior RAS" in my list above as I know that is a nebulous concept often thown around as FUD.

OK, so in the interest of learning more about Linux, help me to understand how I would accomplish moving say an Oracle database from say a pair of 32c vPars in a Serviceguard cluster to Linux...

-I want to keep the ability to virtualize as I might want to add/remove CPU/memory to the database online
-I don't want any virtual IO penalty
-I need to keep High Availability
-I want to manage the CPU/memory resources assigned to this service and to other ones across my whole estate
- The HW/SW has to stay supported and "upgrade-able" for 5 years

I'm not trying to have an argument here - I'm genuinely interested in the opinion of someone who has gone through this - every time I look at this I see issues for my customers, but as I'm happy to restate, "I'm no Linux expert", so I guess I could be missing a trick.

Duncan

[Kenan Erdey]

> I deliberately didn't call out "inferior RAS" in my list above as I know that is a nebulous concept often thown around as FUD.

it's more meaning full, athough a processor is broken and OS and HW will overcome this error without downtime in work hours, in payment or CRM system etc.

by the way you can say i will form a active/active cluster farm with x86, keep the availability without paying more for RAS.

this is the intel's ras features explains xeon,itanium etc..

http://www.intel.com/assets/pdf/whitepaper/ras.pdf

[Alzhy]

Kenan... there really outh to be no need to even consider clustering to provide HA and that mythical "RAS" to say Duncan's current Active/Passive (and expensive) 32-core/XXXGB RAM/ nnxFC Database environment. Although HA solutions abound - like ServiceGuard clone RHCS which most SG afficionados will feel at home fairly quickly and of course the venerable Fortune 50 standard -- Veritas CLuster Server.

Why? Virtualization and Cloud practically negates the need for clustering.

But in case of Duncan's predicament - it is still possible to have 2 Virtual Linux Servers clustered -- for uber high availability (I actually do this too --- 2 Clouds/Virtualisation Clusters with Virtual Servers clustered accross).

Duncan - I come from a somewhat similar HP-UX database environment. Large vPARS on SuperDomes (PARISC 8800/8900). Although our DBs were not SG'd as we employ poor man's clustering. Our DB vPar servers are 12 to 64 cpu vPars and average ~ 160GB of RAM and 8x2Gbit FC Channels. Each we've successfully moved to older generation Dunnington Systems (Xeon 7400s -- 6 core pre-Nehalem X86 Servers). On the average 32 core HP_UX/PARISC environment -- we were averaging ~80% CPU Utilisation. On a 24-core, 4 socket X86-64 Dunnington Physical Linux Server - we are averaging just ~30% CPU Utilisation - that we think moving these environments to either an vSPhere or KVM hypervisored HA Virtualisation CLuster will wikely be possible.

Everyone's likely question will be -- for such HUGE DBs on a virtualized Linux Server - hwo do you ensure there will be no Virtualization OverHead specially I/O?

Well -- for both vMWARE and KVM -- direct SAN or HBA access has been possible. On Vmware environments -- even RDM (raw device map) is already sufficient. What makes this possible? NPIV technology .

As to the CPU and Memory HotPlugging -- there is virtually no need for these features as vSPhere/KVM hypervisors support over commitmment of CPUs and Memory so there is never really any wastage of these reources. But I believe support is there and quickly maturing if ever.

BTW, we used VxVM/VxFS 5.X's CDS features to move the Oracle TTS data over to Linux and used RMAN to do the migrations. Your DBA should be familiar with the process - if not, there should be available Oracle documentation on the process. We were at ASM storage when we did out migration so it involved some conversion on teh DBAs part and fscdsconvert over a floatable VxVM diskgroup to float over and endian convert the data.

Painless - relatively. Only major issue we had was when we moved to 11GR2 from 10GR2 and we were caught unaware by the lack of HugeMem pages as wel as tweaks to swappiness and vm.dirty tweaks which I posted in the Linux threads.

[Duncan Edmonstone]

Alzhy,

OK... so 2 comments/questions:

- VMware and KVM - both have "issues" with Oracle from a support perspective
- Oracle licensing - if I get round my lack of online add/delete of CPU/memory by having "extra" cores free in my system which my VM can call on - I have to pay the Oracle licensing for them, even if I never use them... do you have any way round that?

Cheers,

Duncan

[Alzhy]

1.) Be sure you are able to reproduce your issues ON a Phsyica Server. Or be ready to V2P.

2.) I don't know the current state of Oracle Licensing on a HyperVisored Server - it is likely licensed per physical node though

Overall -- Oracle on Linux is still considerably cheaper.

[Duncan Edmonstone]

Sorry, doesn't work for me... I have to keep another system around just to confirm issues on? That costs a bunch and will significantly slow down analysis... maybe your customers are less risk averse than mine, but there's just no way mine would go for that.

...and yes, Oracle _only_ recognise an x86 VM as a hard partition for licensing if you use OracleVM (i.e. give _all_ your money to Oracle!), so I have to license all the cores on my x86 server, whether I need them or not

- the (obvious) result of this is lots of smaller servers to run individual databases, which a) creates a management headache and b) isolates unused compute resource

so whilst I can see a strong case for what you discuss on a "per project" basis, when I think about this across my customer's multiple datacentres with Oracle instances measured in the hundreds, I can't see how this would stack up.

Duncan

[Kenan Erdey]

conversation turned out to use oracle on linux or unix. partitioning is weak in x86 side and can be done with virtualisation. There is no hard partitioning or partitioning techniques without loss of cpu power, I/O latency, bandwith etc in x86 side.( i know some improvements in kvm side) i didn't try rdm but in integrity VM side, although availability of AVIO, when i tested, physical and disk given from AVIOA didn't give the same results.

We had also started to move linux side for some small databases. An because the penalty of oracle support, What we did is physical consolidation of databases on two active/passive linux nodes.

[Alzhy]

Dunc.. the "be ready to prove issue exists in a Physical Server" approach is just insurance sir - just in case Oracle support "insists" in trying to replicate the issue.. To tell you the truth - we've never been questioned by Oracle what virtualization layer we used -- Pls refer to the following URLs too:

http://www.vmware.com/solutions/partners/alliances/oracle-vmware-support.html
http://www.vmware.com/support/policies/oracle-support.html

I too was skeptical with the Linux "resource management philosophy" having been schooled in nPars and vPars (HP-UX) and Containers/Zones/Domains (Solaris) and their high avaialability suites. Hope your issues, skepticisms and doubts will be addressed as you continue your journey if you decide to or stay with the dinosaurs.

And as I have been espousing in the forums -- it's quite easy and CHEAP for fledgling Linux practitioners to dabble and wet their fingers in Linux HA and Virtualization. If you already own a dual or a quad core (heck even 6 core AMDs are cheap) - have suffcient RAM (mem's cheap too!) and Storage (need I say more?) - then build yourself your own Linux Vortualization lab and try varius Linux, Windows, appliance technologies as easy as 1 2 3. Dabble in NAS, HA CLustering - various LINUX Heads for NAS, Email Gateways, Firewalls, etc. And all are "FREE" -- for hypervisor -- you can opt for Linux's own -- KVM, or Vmware Server hosted on a Linux distro, or bare metal ESXi or Hosted Vortual Box.. heck even Hyper-V if you want to do the 2008R2 test drive.

What I am saying is it is SOOOO easy for systems people to keep up and keep abreast with today's open technologies -- and that's what's killing UNIX.

[vtmue]

Hello.

I'm so tired about reading 'race to the bottom'(1) propaganda like above that I finally registered to this forum.

IT has some specific rules that make it somewhat unique. There are few who have the intellectual capabilities to cope with the inherent complexity of IT in general.
Seldomly these are decision-takers (I'm one, so I know what I speak of, plus I count myself to the stupid majority). Therefor, there has always been a deep desire for having a dramatically less complex IT infrastructure.
This was demonstrated in several scales for decades now, starting with the unparalleled success story of an operating system from Redmond, and continues in other iterations of the dance around the golden calf. Take virtualization for example. While it truly is a big step forward, it is still immature(2) and not at all suitable for all cases it is praised for.
What in my opinion is quite unique for IT is that the race-to-the-bottom appears to go faster and in a more dramatic way than in other industries. Moving from HP-UX to Linux fits as a perfect example.
Everyone who persists in stupid generaliasations like 'running your IT on linux is cheaper' is either a dreamer or doesn't see the whole picture:

The whole picture is that Linux does has numerous fields where it is indeed the most cost-effective player in. These are primarily small to mid-sized, non-critical setups. In any other context any _real_ comparison shows that Linux is on par or worse with regard to TCO. Among the reasons for this are less reliable/scalable hardware base, higher speed of development, shorter support lifecycles, dramatically richer featureset (it is the nature of software that bugs come as a certain percentage of code extent), testing focus on feature popularity instead of significance.

A dreamer is someone who blindly follows other people's tunes without considering/opening their eyes first. So someone who still believes in the future of free operating systems in enterprise context. Someone who keeps repeating how perfect and stable cheap solutions are. Someone who ignores the universal rule that you always get what you paid for. If the latter wouldn't be true, we were still sitting around campfires with our clubs and furs.

In more than 15 years of IT I have been many times feeling the pain when initially paying for Unix systems. I'm looking forward to that kind of pain again. Why that? Because I have also listened to people explaining to me outages of cheaper systems.
E.g. why a defect in SuSE deleted the default route, why a defect in SuSE made NFS mount stop before STP was ready, why OpenBSD froze when VLAN tagging was used, why Debian didn't honour nsswitch.conf settings, why redhat crumbled into pieces while upgrading. And since someone boldly dared to mention vmware not only being on par but a successor to clustering, I simply recommend taking a look into their amazing defect ("knowledge" ;-) database. They just gave "never touch a running system" a whole new quality.

Regardless, for some people "reality" will always be what they want to see.
And, I have to admit, this can pay off if many join forces in doing so over a significant period of time. Consider a server OS from Redmond, which after 18 years of improvements today really is a solid and reliable piece of software.

Maybe /they/ will port their enterprise DB to HP-UX, who knows.

For me, HP-UX still still makes a good choice in a certain scenario, despite the ongoing childish war between a few way-too-rich men at the top of oracle and hp - the former apparently giving a sh*t on customers demands and needs.

My $0.02
vt

1) http://en.wikipedia.org/wiki/Race_to_the_bottom
2) Everyone I know utilizing vmware officially tells how pleased they are, in private then they admit having been bitten by the same rate of defects than we were in our own setups, and that savings sometimes were minimal or simply non-existing.