Linux vs W2000 Server vs Unix
Posted 24 July 2003 - 09:36 AM
Our early (fairly primitive) benchmarks show that Linux works faster than W2000 server but when compared to a Sun Solaris server installation it does not fare as well. Unfortunately we're not comparing like for like as regards chips/memory/disk speed etc. so the benchmarks are not really representative.
Question : Does anyone out there have any experience of porting ProIV 5.5 to a linux environment?
If so what sort of performance and reliability have you seen compared to your original installation.
We are not limited to Linux for this project but scalability and flexibility are issues - we need a solution that will work for large clients and very small ones. Hence the reason we're looking at Linux.
Also, we've seen a couple of teething problems with this release (e.g. SQL embedded SELECT statement with primary file in 'D' mode deletes all records on the file - not just the ones selected by the SQL!!). Does anyone else have any biggies like that we should watch out for?
Thanks in advance
Posted 24 July 2003 - 10:05 AM
We don't have any experience with Linux since none of our customers are using this (yet), but have systems running on Wintel, Solaris and AIX servers with Oracle 8.1.7.
Usual stuff - I would strongly recommend that you test using your application with similar numbers of users and realistic (production) database sizes.
What is the profile of your app, max number of concurrent users and any batch processes, size of database?
FYI we have 400 users running happily on a Solaris site and over 100 on a Wintel site. Note that I think there is a 2Gb limit on Windows for the size of any single process and this will limit the maximum number of users. How scalable do you need it to be and how significant is the hardware cost compared to everything else?
Posted 24 July 2003 - 12:33 PM
My take on this:
You have one major constraint - just how much will the customer pay for the "whole enchilada". If their budget is small, then your choices get more limited. Scalability does need to be factored in especially if the customer expects business and eventually the number of users to grow. They would be poorly served ifthe box they bought into could not be expanded much.
I am seriously trying to move our company from an old proprietory Unix box running YT-100 style terminals into the Windows world. This move world provide the ability to improve desktop performance by expanding (what is in) the box or getting more "horsepower" (by getting another box). I do not want to be tied to a proprietory solution in the future, and forced to replace the entire system again. Your customer needs to give you their perspective; after all you are a "solution provider".
Edited by Bob Filipiak, 24 July 2003 - 12:36 PM.
Posted 25 July 2003 - 07:44 AM
And Windows isnt proprietary?
Seems a bit odd that this is your driving force, and yet you are recommending the leap to windows. What did someone once say about the frying pan and the fire?
If you really want to avoid being tied to a single vendor, have a look at linux. Linux is linux, so if one vendor doesnt perform you are always open to migrate (painlessly) to another vendor.
Edited by devnull, 25 July 2003 - 07:45 AM.
Posted 25 July 2003 - 05:39 PM
I do not want to look like one of Saddam's sons, all bloody and bruised.
OK - Windows is proprietory, but HARDWARE running it does not have to be. What I did not mention was that my aversion was being stuck with a single vendor's HARDWARE. The Unix box I am referring to was made by Convergent Technologies, which somehow got acquired by Unisys. But as far as Unisys is concerned, it is an orphaned child. No support, no information. Parts for it are hard to come by, and get expensive. It uses the original SCSI drives; and the one currenty in use is over 6 years old. At least in a PC world, EVEN if it is running Linux; if the hardware dies, repair is REPLACEMENT. Compare that to the lower end proprietory boxes.
Posted 30 July 2003 - 06:52 PM
You might be looking for something that isn't possible.
We have a complex application running under UNIX and sold a port to Linux but we then had problems with performance and PROIV internals so we had to pursuade our clients to pay for more expensive (UNIX) hardware.
Tell them to pay up and stick with what works.
Posted 31 July 2003 - 10:11 AM
Thanks for the feedback on your experiences with Linux. This is exactly the sort of feedback that i was hoping for. We've been on Linux 7.3/Oracle 8 for a month now and things seem pretty stable. What versions of O/S, hardware and database were you running on ???
I way I see it is that Linux is essentially the same as Unix. But in our ProIV world, we are limited in that we have to run on intel-based hardware. Therefore we can't benefit from the performance advantages of faster servers as you would if you selected say, a high-end HP unix server.
I wonder if ProIV are considering supporting Linux on other chips. 3rd parties such as IBM/HP etc are already providing Linux-based alternatives. Are ProIV going to follow suit and support not-intel Linux environments ???
We are considering Linux because it is scalable and enables us to have a top-end solution and a low cost solution on the same O/S. This is important for our target market.
but thanks again for the feedback - does anyone else out there have bad experiences with ProIV/Linux ???
Posted 31 July 2003 - 11:08 AM
I believe ProIV are preparing to support Linux on the IBM Mainframe and will support Linux on Itanium (the Intel/HP IA64 architecture) as soon as there is sufficient demand.
HP's proprietary chips are at the end of their natural life-cycle and Itanium is the future for HP-UX, OpenVMS and for Linux from HP and other vendors.
Linux is capable of supporting large multiprocessor machines nowadays (and I think Oracle clusters now too) so it should be possible to achieve very high performance levels with Linux.
Note problems with SQL SELECT and similar in ProIV are very unlikely to be Linux-specific (although I guess it's possible).
Posted 31 July 2003 - 11:21 AM
The only problem of course with Oracle is that while Linux, and the hardware, is remarkably cheap (try a 32-blade cluster for under 100,000 USD), Oracle is still pretty much at the expensive end of town in the RDBMS market (undiscounted it's about 40,000 USD per CPU).
We were considering RAC for a while, we may even go there in the future if business continues to grow at the rate it's been growing, but for now it remains an expensive option, even if the o/s and hardware is cheap. RAC appears to scale extremely well, and Oracle promise that (with 10i?) they'll support it on pretty much any platform, clusterware notwithstanding on the o/s.
Posted 31 July 2003 - 04:21 PM
The STANDARD edition is $15,000 per processor OR $300 per named user.
Only the ENTERPRISE edition is $40,000 per processor, when licensed that way.
There are also licenses with 2- and 4-year terms at reduced prices.
The minimum number of named users for the standard edition is 5.
The standard edition can be licensed on a named-user basis for hardware with up to 4 processors.
So you can have Oracle database standard edition for 5 named users on a box with up to four processors for $1500 (this is a perpetual license).
More detail available at http://oraclestore.oracle.com
Posted 01 August 2003 - 07:18 AM
for most real applications. If stuff like replication (and I guess the
cluster s/w) is needed then the enterprise edition would be required.
This is why, I guess, SQL server is so common now, because Oracle
is way too expensive. Despite what their sales team say...
of the poster and do not represent those of any organisation.
Posted 01 August 2003 - 11:07 AM
Basic replication IS included in sthe standard edition. You only need the Enterprise edition for Advanced Replication (eg. multi-master replication with rule-based conflict resolution). If you really need that then I think you are in the "enterprise software" league and would expect to pay those prices, no?
Real Application Clusters do need indeed Enterprise edition (and may even be extra cost on top of that) but I think the same point applies - why would you need that unless you have thousands of users and/or need 24x7 availability?
Maybe Linux clusters will bring down the RAC price point, we'll see.
Reply to this topic
0 user(s) are reading this topic
0 members, 0 guests, 0 anonymous users