Pa-Risc or Itanium

[Michael Denney]

Hi all,

My company is planning a huge project. The plan is to buy some of the hardware now. The plan is Design phase in 2005, Implementation and Testing in 2006. Go production in 1Q 2007. The application requirement is a 4 tier application. -->> Database server(s)--App server(s)--Web server(s)---browser. Given that the application was certified for Itanium in summer 2004 and the database is Oracle RAC..what platform would you recommend? We see the choices as A. PA-RISC B. Itanium2 C. Go to a new vendor (Sun or IBM). Please give informed opinions. Thanks so much for your time. Regards,Michael

[Michael Denney]

Appendum to the needs:

1. Thsese will be new business critical servers.

2. High Availability and performance is a must

3. Cost is also a concern, but secondary to performance and High Availability.

[Patrick Wallek]

Hi Michael,

I stil favor PA-RISC, but Itanium is the way of the future.

I personally think that you are jumping the gun here. With the design phase of the project not starting until next year, I think it is a bad idea to buy hardware now. If you do that, you could get into the design phase and discover you bought something too small, got the wrong options, can't have enough RAM, etc......

I would wait until the design phase of the project is underway, then talk to HP, or an HP partner adn get them involved and let them make their recommendations.

Without knowing the amount of data, response time requirements, the number of people using the application, and a number of other factors, it is very difficult to make a good recommendaiton right now.

Don't try to shortcut the process. You'll regret it later if you do.

[A. Clay Stephenson]

I would go with Itanium given your timeframe for production. Now I'm going to throw you a real curve. There is actually an advantage to buying somewhat under-powered test and development boxes. My reasoning is this: Slow boxes are great for spotting poor algorithms and data structures. The last thing you want is for test/deveopment to seem like its great but when the volumes ramp up, the production environment slows to a crawl. Developer's hate my approach saying things like compiles take forever. The truth is that compiling/linking might represent 1% (and that's a very generous fraction) of their total development time. You could make a good case for buying used equipment now (PA-RISC or Itanium). The code should port just fine to either environment if your developer's are disciplined.

The other thing to think about is that this doesn't have to be a homogeneous configuration. Your database server(s) might be HP-UX; you should strongly consider Service Guard. You application or web servers might be Linux; again, consider MC/SG. All of this should be monitored with something like Vantage Point/Operations (or the current product name of the month for this Openview product). You should also give very strong consideration to what is often the last examined task of a project like this: backup. Your backups might be extremelt complex if you have database and application metadata which must be kept in sync.

[Sridhar Bhaskarla]

Hi Michael,

How are you doing?. Congratulations on your new project.

I was going through this white paper sometime back. This may be a slightly old paper but gives excellent performance and cost comparison between HP's integrity servers and other vendors. It's very interesting to see HP integrity superdome with 11iv2 doing better than IBM and PARISC superdome.

http://h71028.www7.hp.com/ERC/downloads/cost_ownership.pdf

Since the project is going to start in 2006, I would tend to go with itanium servers as HP says that they are the future of HP. But it doesn't hurt gathering information about other vendors too. Since the project is huge, I imagine you will have a number of systems in your datacenters. So, you may bargain with other vendors (Sun and IBM) to see if they can setup some of their systems for 'proof of concept' tests.

Also, HP has started offering 'pay per use' type of solutions so you don't need to buy the equipment and you pay for only what you use.

http://www.hp.com/hps/utility/

You can also consider the above.

-Sri

[Michael Denney]

Hi Sridhar,

Doing good, thanks for asking. Quite pleased to see your opinion here Sri. I always enjoy reading your theories here.

A bit more about the project. We are not purchasing all the hardware up front. We are making a purchase for the 2005 design stage. We envision a sort of mini-production environment. We will purchase more hardware later in the project.

Yes, we are going Oracle RAC-MC/Serviceguard on the database. Yes, I've been considering a mixed environment.

Considering: Databases run on HP-Itanium due to performance/cost, App servers run HP-PA-RISC due to application certification and web servers running Linux due to price/simplicity of web servers. Also seriously considering going to AIX due to the unclear strategic roadmap of hpux/pa-risc/itanium.

I've been doing hpux for 7 years and don't know AIX at all. However, I'm considering AIX as "what is best" for my company and this project.

Thanks much for opinions. Keep em coming!

[H.Merijn Brand (procura]

I'm now using AIX and HP-UX for almost 10 years side by side, and I still prefer HP-UX 10.20 over AIX 5.2

In my perception, and what I experienced with clients, HP-UX is much more stable than AIX.

AIX has only two major advantages: the LVM is the best I've ever come accross (note that I have never used Solaris), and smit is way easier to use than sam.

Though AIX is already much better than Dec/OSF or Tru64, it's still a pain to remember the unlogical commands they use for system administration. If you onloy use sam/smit, you probably won't notice, but in times of trouble, you will have to revert to the command line eventually.

I'm also a Linux user, and HP supports Open Source *MUCH* better than IBM. Try to find *recent* gcc builds for AIX, or a working GNU make. You'll be in big trouble if you need it to make HP-Linux-IBM connections that depend on open source projects (Samba, OpenSSL, OpenLDAP, OpenODBC). Though I must admit that building those utils on HP is not always as easy as the docs make you beleive, it is very well doable, as my site and the HP Porting Center prove day after day. Open Source projects also build easier on Itanium than on PA-Risc, but I was until now not impressed by the performance of Itanium systems (I only used HP Testdrive systems, which might not be optimally configured for the tasks I used them for)

These are personal opinions and reflections of my experiences. I hope they help on your decisions. Good luck.

Enjoy, Have FUN! H.Merijn

[Geoff Wild]

Sun would be the last I would choose - I have worked with just about all flavours over that last 11 years; HP-UX, AIX, Solaris/SunOS, Irix, Linux, BSDI, Ultrix.

HP-UX is by far my favourite...

AIX is close...

For your project timeline, with PARISC platform winding down (unfortunatley) - I would go Itanium 2....I know we will be - probably our next purchase will be Itanium 2...

Depending on your Oracle licensing, you could save money with Itanium - as benchmarks show you can do the same with 2 Itanium processors as 4 parisc's.

Stick with HP - support is far superior then the other vendors...

On top of that, no one can even come close to these forums. :)

Rgds...Geoff

[Steven E. Protter]

HP is the most reliable hardware/os combination.

To decide on PA-RISC/Itanium, take a few factors into account.

1) Applicationis. I have application vendors that quite openly say they don't support Itanium, even HP-UX on Itantium. Until they support it, we're a PA-RISC shop.

2) Long term growth and support. If you don't buy systems very often and your applications are supported, HP has made it very clear that Itanium is the supported platform. There will be processor bumps and upgrades and there is a longer life cycle.

SEP

[Alzhy]

PA-Risk or Itanic?

The PARISC line is a safe bet if your technology refresh cycles are in 5-6 year periods. The PA-8800 (dual core) was just released last Spring and will be on its last update with the PA-8900 - ? years from now.

The Itanium2 line certainly is the fastest HP boxen right now.. currently breaking records here and there but the problem is only HP is actively pushing this CPU at the Enterprise level.

I respect HP-UX as a really stable OS specially on PARISC platforms but I suggest you widen your horizons and also include STORAGE in your plans. Also, do ask for test platforms and do actual tests of your applications.

If you're a really BIG enterprise and stability is of prime iportance over performance and cost is secondary - stick with PARISC + HP-UX. Otherwise, if you're adventurous, cost conscious and have in-house wizardry - go for a cluster of Quad CPU Opterons or XEONs running UNIX (Solaris 10) or Linux. Man it would have been nice if HP-UX can run on non-Integrity Itaniums and the Opteron/86-64 chips...

[Ted Buis]

The intestment protection is very good between pa-8800 systems today and Itanium 2 based systems. They use the same physical boxes, chipsets and memory. I would recommend Itegrity (Itanium 2) based servers for the Oracle DB (RAC or 10g). The application servers might need to be pa-risc today, but remember support is for 5 years after the end of production for hardware. Remember that you do have similar issues with different releases of hardware (Power 4 vs. Power 5) and AIX 5.2 versus 5.3. Very few applications are available on AIX 5.3 today, but to take advantage of Power 5 you need 5.3. If you look at the investment protection in the past for HP-UX, Superdome for example has been able to be upgraded from 550MHz to 750MHz to 875MHz to 1 GHz dual core, to Itanium 2 and future Itaniums while IBM has gone from the 3 generations of totally different hardware on their high end. On the SUN side, SPARC V is dead, as they go to the Fujitsu chip, then back to their own chip. No compatibility with those systems looking forward.

[Kent Ostby]

I work for HP so I'm obviously not going to steer you to another vendor.

On the question of PA-Risc versus Itanium, I would somewhat ask myself the question of how much this project might expand into the future.

In other words if you can see the maximum size (barring completely unforseeable issues like you're company developing the next "big thing") then I think you can go either way.

If this project is, for instance, the first step in doing such and such and we plan to keep rolling it out to the rest of the company for x number of years then I would go with I2.

Oz

[Ted Buis]

Another point, IBM has no hardware partitioning, and SUN has no software partitioning. HP has both hardware and software partitioning for greater efficiency.

[Tim D Fulford]

Hi

PA RISC is good, but Itanium is better. HP say that a dual core PA8800 will do a similar amount as a sing;e core IA64. The MX2 chips (dual core IA64) will be coming out by the time you go live & the chassis for the Itanium2 will be compatable with the dual core versions (according to the road maps I saw).

Licencing. Oracle etc are licenced by how many CPUs you have. So sticking to PA-RISC will, in-general, mean you will double your database licencing costs.

Switching to IBM etc, cant really say..

As you are probably aware the H/W scene will have dramatically changed by 2007. IA64 is new & looks likely to grow, PA-RISC has only one more chip available on the roadmap (PA8900)... BUT if you do go PA-RISC route you should be able to switch to Itanium system provide you use latest H/W for PA8800.

Tim

[G. Vrijhoeven]

Hi Michael,

One advantage of Ithanium over PA-risc is the speed, pricing and roadmap perspective. Since you will test the config first and the middleware is supported i would go for ithanium.
Option C SUN/IBM. If you have the experience on all platforms and monitoring agents/backup agents/work instauctions are availible for multiple OS configs why not. I I would also consider linux for the webserver.
Regards,
Gideon

[Ted Buis]

This is really in response to Nelson, saying it would be nice if HP-UX would come out on Opteron. If it did, it would suffer from the same problems that Solaris on x86 suffers, which are that there would be no binary compatibility and worse, x86 is little endian and pa-risc/sparc are big endian, so there would be no simple data compatibility either. Itanium is bi-endian so it can go either way.

Also another alternative would always be LINUX on Itanium. Oracle has posted impressive TPC-C numbers with Oracle 10g on HP Integrity servers.

[Alzhy]

Ted,

First on the Partitioning Offerings of the "Big 3".

Until this year, HP probably had the best partitioning spectrum on UNIX -- from Hard Partitions (nPars - which in itself has some importantlimitations. ie. no Dynamic Reconfiguration) to "soft partitions" or vPars (which is at the CPU Level) and finally to PSETS (processor sets - which coupled with WLM - allow you to "effectively" utilize your CPUs. With ever powerful CPUs - it has become evident that sub-CPU partitioning will become important and this is where IBM is I think solidly in the lead as AIX cannow run in micro-partitions instead as the whole CPU as the smallest partitionig unit. SUN with its Solaris 10 will offer a totally divergent approach to virtual partitioning via "Containers" - wherein instead of separate OS's. each container will be a self contained operating environment for even more effective resource utilization.

I do not favour any vendor - I simply am practical with my choices and I am for CHOICE - that's why Linux is so great an OS becuase it gives everyone the power of CHOICE.

Now on to Solaris 10. Yes Sun has different verions for SPARC, x86 and X86-64 but it offers the customer CHOICE. I've previewed this OS and this release and offers "promise". It's "containers" approach to partitioning is totally radical in that it did not follow the inductry trend of providing a separate OS for each "container". It's dTrace utility will totaly remove the guesswork out of the chronic question - "Why is my Application Slow". It's upcoming Filesystem (zfs?) will reportedly be Endian-less - so if an Opteron environment can no longer provide the processing power.. simply mount the data over to a bigger Sparc machine - thereby eliminating the need to rebuild environments.

You probably have noticed in my first reply: PA-Risk or Itanic? I withdraw that - it should now read: PA-RISC or Itanic. So for Michaels dilemma - I would still go for a PA-RISC system - if eveyone still believes that after the PA-8800, HP will still have enough dinero to reach the last release - the (quad-core?) PA-8900.

[Ted Buis]

But IBM's sub-CPU partitioning doesn't support HACMP and their LPARS "support" of HACMP leaves much to be desired, as reconfiguration of resources can cause suspension of activity that takes enough time that it can be interpreted by HACMP to be a down partition. HP was the first vendor out with software partitioning, and vPars have better isolation and lower overhead than LPARS. Since they are different designs, there are trade-offs, but I wouldn't be so quick to say IBM's sub-CPU partitioning is better. Also, of interest Oracle has now said they support vPars for licensing considerations. Further, have you read about the Integrity Virtual Machine targeted for late next year. While it provides sub-CPU partitioning, it is not designed to replace vPars. Each approach has its strengths and limitations. With vPars you have excellent isolation between partitions. It is here today with much application software. IBM's sub-CPU partitioning requires AIX 5.3 for which there are few applications qualified.