24x7 hpux oracle environment ....

[Henrique Silva_3]

Sorry for the x-posting ( similar message on white papers and database forums ).

I am starting to investigate the implementation of a single oracle site ( probably two superdomes with service guard and/or replication on a single site ) with app servers on the same subnet connecting to them, and clients coming from all over the globo to it ( about 100 k users ).

Need to know anything about what archtecture should I use, white papers, redundancy, storage ( XPs ), etc. I am also, looking into a second scenario, where we would have several servers on the different regions of the globe, replicating the data.

ANy help here would be appreciated.

Thank you much,

Henrique Silva

[Patrick Preuss]

hi,

some questions to think about:

the traffic producing over the WAN Links in the senarios?

the max failover time?

witch data need witch data and wehere do they resist?



best regards

patrick

[Tor-Arne Nostdal · Accepted solution]

Hi
We are currently running such a described solution - though we dont have the same amount of users (defined 5-6.000, normally not more than 1.500 simultanious).
Global VPN network + other network connections.

We're using running only one DB-server (superdome), but have failover to another (separate superdome), XP512 disk storage, tape library, and a number of application servers with load balancing for the application (SAP R/3). Remember to net everything together with enough bandwith both on cables/switches/routers aso.
Our Oracle DB is approx. 450Gb
We have several copies of the system as well.
Full copy for:
- Development (with scrambled HR data)
- Test
- Separate project excecution systems (example when introducing Euro)
etc.

MC Serviceguard is a good product to achieve high availability, but also remember backup-time.

We have been scaling this solution several times, changed from EMC to XP512 etc. but still have maintained a very high operability.

Today we do Offline backup, split-mirror backup once a week which takes approx. 15/20min. downtime (nobody want to guarantee that online+archive logs is enough - though we never have had to go all the way back to this).

We will continue a very centralized policy, but we put down a lot of effort in preventive actions. This is done both by 24x7 standby duty, monitoring tools with messaging systems, auditing/benchmarking several times a year etc.

Though we have a centralized implementation - we are a de-centralized team of system managers.
This requires also good documentation (exactly where the machine is located and where the power button is, as well as a streamlined computer room with everything labeled).
We can start/stop Diesel generators remotely, check UPS etc. (but we cant refuel :-))

What we now have on our wish list is a second computer room (approx. 2km away). Then we can move half our implementation and also stand a fairly large bomb :)
Today we have an agreement with HP (Business recovery solution) - where we can bring up our system within 72 hours (in another european country).

There's a lot to consider - but remember that the most vital thing is to make sure you have something you can grow in...

/Tor-Arne

[Henrique Silva_3]

Thanks for the "extensive" post :-)))

That is exactly what I am looking for. I do not think our database will be that large. It will probably fit into ram. The app servers will be ALL in the same subnet, so, the only network traffic will probably be clients from all over to this site.

I am looking at it from an oracle perspective right now.

I will probably have two superdomes, with service guard and/or advanced replication going between these two servers, on XP512s and maybe 50 GB to 100 GB.

I am looking at the impact of replication ( latency time, blocked tables ) on this database when multiple users are hitting it.

I think the user base is large but we would not have more than 10,000 concurrent user at the time here.

Do you know of any specific white paper that would describe such implementation, or do you have any docs yourself that you are "allowed" to share ? Published ???? I have looked at metalink and I can see some white papers there, but nothing too specific !!!

Thanks,

Henrique Silva

[Anonymous]

Tor-Arne,
do you mind sharing the
failover time of your oracle DB
from nodeA to nodeB using MC/SG
with us? ...just curious...
tia
Thomas

[Henrique Silva_3]

Tor-Arne.

Are you using SG for failover, correct ? And ONE XP512 or two ??? ( Single ppoint of failure would be the XP here)

If you have two XP512s, how are you replicating the data for the failover DB ? Advanced Replication ? Standby DB ? HPUX mirroring on the XPs ??????

If you are using advanced replication, can you comment on any benchmarking you have done ? I am thinking about setting up two-way replication, but I am concerned with the latency of transactions and any synchronization issues. !!!!! Maybe one-way replication would be safer.

I contact oracle and they do not have ANY benchmarking on replication :-(((

Thanks,

Henrique