Operating System - HP-UX
1748074 Members
5425 Online
108758 Solutions
New Discussion юеВ

Re: Performance - Oracle 9i on different RPxxxx platforms

 
SOLVED
Go to solution
Carl Houseman
Super Advisor

Performance - Oracle 9i on different RPxxxx platforms

I'm operating a 9i database and Oracle apps 11i on a a 2-cell-board RP7420 that has 24GB and 9 900MHz processors allocated to the production vPar. The actual processor loading on this vPar rarely exceeds 4 processors' (4 cores) worth. The storage is a VA7110 on 2GBPS A6795A FC cards. The typical load characteristic is I/O limited - we got a noticable bump in performance when the VA7110 was re-done from all data in RAID 5DP to all data in RAID 0+1.

We are looking at acquiring a "hot standby" box to keep in sync with production and use for emergencies when the RP7420 is unexpectedly down for more than 2 hours. The platform considered is an RP3440 with 4 x 1GHz cores (2 modules) and 24GB of RAM. Same A6795As and same VA7110 configuration as production.

I know the RP3440 has only 8.5GB/s memory bandwidth compared to 16GB/s for a single cell board in the RP7420. That seems to be the major difference. The question is, given the I/O-bound nature of the load (most of the time), the 11% faster cores, and no vPar overhead in the RP3440, can I expect it to carry the load well enough to keep people from complaining too much?

thanks all...
9 REPLIES 9
A. Clay Stephenson
Acclaimed Contributor
Solution

Re: Performance - Oracle 9i on different RPxxxx platforms

I think that this will be a perfectly acceptable fail-over box and because they have the same amount of physical memory, you should be able to run identical database tunings. Your users may not actually notice any differences in response but that is one of those "it depends" areas. My rule of thumb for "limp-along" boxes is that they must be able to run at about 60% of the "good" box --- and you should have no trouble meeting that. You simply need to explain to your management/users that you are trading dollars for possible limited periods of performance degradation. Because of the overhead of keeping the standby database "in sync", a Plan B immediately suggests itself. Why not use Service Guard? Your fail-over node could then automatically takeover seamlessly (assuming the clients can handle the brief transition period).
If it ain't broke, I can fix that.
Carl Houseman
Super Advisor

Re: Performance - Oracle 9i on different RPxxxx platforms

Hi Clay, thanks for the feedback.

I'm not a DBA, just the Unix guy, so some of these Oracle things are not familiar to me. You have said "service guard" which I assume is an Oracle thing, but our DBA's have talked about configuring a "data guard broker". Care to give 30 seconds to 'splain the difference, and maybe why we might choose one over the other?

thanks,
Carl
A. Clay Stephenson
Acclaimed Contributor

Re: Performance - Oracle 9i on different RPxxxx platforms

Service Guard is not a dba thing; it is an HP-UX / Linux thing. Imagine that you have multiple hosts (at least 2) that constantly talk back and forth to one another saying "I'm OK; how about you?" (Heartbeat). Now imagine that you have multiple disks or arrays, multiple networks, multiple SCSI paths, multiple power supplies, ... so that the failure of any one thing does not cause an outage. At this point you have reduced the proability of failure greatly BUT still the loss of a host (computer) would cause you grief.

Here is where SG comes into play. Imagine that your clients do not connect to a host computer but rather to a "package". A packager has resources: IP Address, Disks, and Software -- e.g. a database instance. The package IP address and hostname are virtual and can be moved from host to host automatically. Your clients connect to host "donald" but the virtual hostname "donald" might actually reside on physical hosts "huey", "louie", or "dewey". The clients neither know nor care about which physical host they are actually running on.


Just start here:
http://docs.hp.com/en/ha.html#Serviceguard

I went 5.8 years with zero unplanned downtime on an Oracle-based ERP implementation under SG.

SG is also something nice to have on your resume.
If it ain't broke, I can fix that.
Jeff_Traigle
Honored Contributor

Re: Performance - Oracle 9i on different RPxxxx platforms

Serviceguard is HP's high-availability clustering software. You'd have shared storage, the two VA's in your case, between the nodes holding the database files/raw devices. You'd use MirrorDisk/UX to keep the data files/raw devices synced real-time at the LVM layer between the VAs. You would definitely need to buy Serviceguard (and get trained on since you've obviously not used it previously). If you're not already using MirrorDisk/UX on vg00, you may need to purchase that also.

From a quick peek at Oracle's web site:

Oracle Data Guard is the management, monitoring, and automation software infrastructure that creates, maintains, and monitors one or more standby databases to protect enterprise data from failures, disasters, errors, and corruptions.

http://www.oracle.com/technology/deploy/availability/htdocs/DataGuardOverview.html

Data Guard will keep all the transactions in sync between the primary and standby database. You may be licensed for this already. (It wasn't clear from the web page whether it was part of base Oracle 10g or an additional cost option.)

Although SG is a good thing to have on your resume, as A. Clay stated, I doubt seriously that'd be a selling point to management to go that route. :)
--
Jeff Traigle
Carl Houseman
Super Advisor

Re: Performance - Oracle 9i on different RPxxxx platforms

Thanks, I'm skimming through docs now. I see the word "cluster" and I think "complexity". However if it avoids all the complexity of log shipping and reverse failover to return to the primary system, that might be a sellable benefit.

My architecture will definitely be "two data center" with the two locations connected at 100 Mbps. Any ballpark on the cost of admission to the SG club?
Carl Houseman
Super Advisor

Re: Performance - Oracle 9i on different RPxxxx platforms

Reading a little further - it appears I must have FC links between data centers, and that is not possible in my plan. Am I reading everything correctly?
A. Clay Stephenson
Acclaimed Contributor

Re: Performance - Oracle 9i on different RPxxxx platforms

Yes, that is correct but you also need to consider the bandwidth needed to keep your databases in sync. You pay for less bandwidth with higher latency --- so how current must your standby data be? You should also consider the consequences of extremely heavy transaction periods.

One of the main reason to install SG is so that you don't ever use it. You see. by the time, you have nailed all your single points of failure (SPOF's) down, SG itself very seldom comes into play. By the time, you have done this, the cost of the SG software is quite trivial.

Take a very hard look at your network costs because I have seen several cases where it was actually much cheaper to build a new data center closer to one of the existing sites than it was to maintain two existing data centers because over time, network costs tend to swamp everything else if high bandwidth over significant distance is required.
If it ain't broke, I can fix that.
Carl Houseman
Super Advisor

Re: Performance - Oracle 9i on different RPxxxx platforms

Without getting bogged down in the details, I'll just to leave it at "putting the data centers close to each other goes against the reason for having two separate data centers".

Years ago I ran VMSclusters over 10Mbps ethernet. Shame I can't make an HP-UX cluster over 100 Mbps....
A. Clay Stephenson
Acclaimed Contributor

Re: Performance - Oracle 9i on different RPxxxx platforms

Note that "close" by no means in or near the same building. A campus or metro cluster can be separated by a significant distance but still within reasonable range for data replication with low latency at reasonable cost. Even if the data centers are fairly close physically, the WAN connections can be through totally different carriers so that short of a major catastrophe (earthquake, very large bomb, tornado, hurricane) the probability of taking out both data centers is quite low.

Any high availability solution is going to be expensive and database replication and SG are not mutually exclusive. Before you do much of anything, you need to plan very carefully.
If it ain't broke, I can fix that.