How to build an oracle server for a large database?

Question

Oracle

It is required to assemble a hardware and software platform for deploying a powerful federal-level Oracle server. A database of several TB in size and an electronic archive of scanned documents are supposed to be 100 TB according to some data, 400 according to others. Most likely, 6 RHEL will be chosen as the operating system, the authorities put forward HP BladeSystem c7000 Enclosure as the hardware platform.
There are options here on how to assemble it. Take a couple of the most powerful blades into the blade and raise the database of the most "heavy" cities on them, and raise the "light" database on the other smaller blades. Another option is to put the same blades and combine them all with some software into one global powerful cloud, in which all the necessary databases would spin, sharing all the resources among themselves.
I like the second option much more than the first, but it also raises more questions. What software will raise this cloud? How quickly and efficiently will it work? What will be the performance loss compared to option 1? How resistant to failure of single blades will it be?

The other side of the problem is the storage of an electronic archive. It is not yet known exactly what it will be, but most likely it will be either a separate database table or a separate Oracle database. With inexpensive 600GB HP hard drives, how do you assemble such storage with the required redundancy and redundancy? It is the reservation on any media that seems to be the least promising in the matter of such an electronic archive. We need a scheme for saving information without backup. Which?

Answer 1

Contact professional integrators, it will be easier, cheaper and faster. all the same, you are unlikely to be able to set up the hardware, software and oracle itself in such a way that it holds high loads well, does not fall and does not shine at all in the warnings of the monitoring system.
Besides, it is a lot of any reefs in HCL sheets, and the same Dell. HP, IBM have already explicitly solved such problems more than a dozen times, so ready-made solutions for such venerable monsters will clearly take into account both your requirements and those requirements that you may not be aware of at the moment.

Answer 2

I would like to believe that you are asking for educational purposes, and are not going to build a server for such a database using manuals and answers on a toaster.

Answer 3

I support, I think that the blade is not at all what you need. You need 1-2 powerful servers and normal storage. Any system integrator will quickly draw up a TCO for you for such a task. Just decide what OS you will have - if Linux, then there is no point in taking SPARC.

Answer 4

1. Also, do not forget 6 RHEL does not pass the compatibility matrix with Oracle 11gR2
2. It is better to take an external storage system with FC support.
3. Servers should be taken with an eye on the software, to support several databases

Answer 5

I asked in the hope of getting advice and hints from experienced people. So to speak, the direction. A directive came from the authorities - to select the HP BladeSystem c7000 Enclosure config, suitable for our task. I realized that this device fits rather weakly. So, in the end, whatever the federal center decides, we will work on that. Or suffer. Thank you all for your (co)participation!

Answer 6

Let me fantasize a little. If you can’t get away from the blades, it might be worth trying to raise the RAC with all active nodes. The advantage is that it is easier to organize a fast interconnect between the blades. The main problem in this case is to correctly design the database in order to minimize the use of interconnect between cluster nodes. And of course, you can't do without storage.

Answer 7

Are there requirements for fault tolerance, downtime?
If 24x7 is required, then either create Standby for each base, or raise RAC. With RAC, as far as I heard (I didn’t implement it myself), there are a lot of strange dances with a tambourine, but you will get load balancing + fault tolerance (if the application can work with RAC, quickly switching to another node).
Regarding storage - if you do not have High-end storage with good computing power, then it's better to collect 1 + 0; if an accident occurs on 5 or 6, then there is a high risk of the death of another disk in the group while data is being restored to the Hot spare; most likely, you have 600 Gb hard drives SATA and about 7200 rpm, so they will be very slow.
In any case, backups are worth considering, even rare ones; even with good storage and high redundancy RAID, you are not immune from controller death.