Still confused about 50% requirement

[murali_balreddy2]

I am still confused about the 50% requirement for better performance.

According to the Note 917803: "Memory requirement of the main memory: Due to temporary indexes and the memory management unit, the main memory requirement may increase to a maximum of twice the size.",

if my index size is 20GB:

1) do I need a physical memory of 40GB (twice the index size)

or

2) do I need a physical memory of 80GB (twice the index size and taking into account 50% requirement).

If the second one is correct, then it implies 4 times the index size is required, effectively shelling 4 times more for licenses.

Which one is correct?

[Former Member · Accepted solution]

Hi Murali,

Let me explain it with an example. According to the "sizing program", say you need 20 GB.

It means that you will need another 20GB for temp indexes etc ( 50%). So 40 GB equals about to 2 x 24GB blades (48 GB total). This is what you will pay for the licenses. You will also need one fail-over blade. + if you need an explorer accelerated blade, you will need another blade to host that too + consider some growth add couple more blades.

Cheers

Tansu

Edited by: tansu aksu on Jan 27, 2011 11:34 PM

[murali_balreddy2 · Accepted solution]

Thanks Tansu.

In the case of my installation, my TREXADMIN summary says I have

Memory 141.1 GB Phys-Avail, 53.2 GB ( 38%) used

Indexes 27.89 GB memory

So it mean, I have 141.1GB - 53.2GB = 88GB (approx) free memory to be filled.

Assuming the current scenario of 27.89GB index occupying 53GB, Can I still store a max of 44GB in the available 88GB memory free space?

Thank you,

Murali

[Vitaliy-R · Accepted solution]

Murali, the "rule" is simple and it says that you can load 141.1GB / 2 - 27.89GB = ~40GB more data.

What it does not say is when you can load those And what I mean is that you need to keep in mind that your system is live and growing, so you need to have enough space for future growth of your data volumes for the timeline forecasted during the purchase of your BWA.

Otherwise you can load all the cubes you want, but then tell your business "You can analyze data only up to today and you will not see any updates to existing data starting tomorrow because there is no space to load daily batches"

[jgleichmann · Accepted solution]

Hi Murali,

I see that I have confused you with this rule from SAP...

The main things are already mentioned above.

Keep some things in mind:

1) backup blade (for failover, it will also take memory depending on your backup mode)

2) reorg (reorgs will free space and also delete temp indexes)

3) 50% rule -> you can orient on this rule, but I think you should play with it and you can use more than 50%...

Regards,

Jens

[murali_balreddy2 · Accepted solution]

Is there a way to see the memory usage over a period, either in BW (RSDDBIAMON) or in BWA (linux level), like a graph or values of used memory (indextempothers) versus time? We are using 7.0, 53.

[Former Member · Accepted solution]

Helllo

You can use python standalone tool to get graph data, i believe it keeps only 3 days of data, but you can write some script to collect the data everyday. But this would be a snapshot of that time though, it's better than nothing but I am not sure how accurate would that be due to temp indexes at that time. If your goal is to monitor to growth, you can just go with the BWA capacity alerts / you can lower the threshold if you want.

When it comes to where to look at how much memory available, you can also try TREXADMIN \ advanced mode \ hosts \ click to Memory button. The "Total Memory Used" column should match with the one you see under the Summary \ Landscape \ Memory (xx%) used.

I assume what you're trying to do is to see if you have enough space to load more cubes - is that right? If that's the case, it really depends on how much data you have + if the majority of the MD is already shared or not.

Go to RSDDV create an index (don't fill though), than go to RSDDV \ BIA Indexes \ find the newly created cube index \ expend the table indexes. You will see the individual indexes for this table there. Some will be green(already indexed due to the other cubes) and some will be gray. Look at the gray number of records and use your judgement. If you think that is not huge numbers try to fill the indexes in the weekend + reorg, and check the memory. The worst case scenario you can delete it + reorg before monday. I know this is not an "academic answer" rather a practical one.

Cheers

Tansu