cancel
Showing results for 
Search instead for 
Did you mean: 

Choosing SAP Lumira Server HANA XS (LREA) hardware

mike_howles4
Active Contributor
0 Kudos

We are beginning to look at Lumira Server a little closer, so the more recent questions we are asking are around hardware.

I've looked at the Lumira Sizing Guide: (http://www.sdn.sap.com/irj/scn/go/portal/prtroot/docs/library/uuid/60271130-0c90-3110-07a0-fe54fd2de...)

It talks about figuring out how many users and concurrency but today these things are big question marks for us since we have no trajectory or basis to even guess how much or little it would be.  Because of this, any inputs I put into the HANA QuickSizer https://websmp109.sap-ag.de/quicksizer are just shots in the dark.  We'd rather of course err on the larger side of caution than too small.

I've tried Googling and searching SCN for some information, so please send me there if I've missed a good link, but here's some of my questions I'm just looking for any other customers or SAP's perspective on:

  1. What would be an 'average' sized HANA XS install?  256GB?  512GB?  Is that much overkill?  Let's say under 100 concurrent users.  I can't even begin to project the volumes in terms of datasets they'd bring, but to throw a number in the air, 32GB of user data (no idea if that's too high or low)
  2. What's the compression ratio for that data?  I've heard 7:1 been thrown around before and that's actually what quicksizer defaulted to.
  3. I've visited this hardware page and it talks about Single Node vs Scale Out configurations.  Any recommendations or limitations (or links where this has already been answered) for Lumira Server?
  4. Cost-wise, I've visited this link from March of this year by and and I'm wondering if these sample configurations would be valid for a decent Lumira server and if the prices have fallen even more?
  5. Who usually does this sizing work?  Surely not JUST a BI resource, does Basis usually get involved at some point to help?
  6. Any hardware vendor recommendations or experiences?  Feel free to PM me instead in case you don't think it's ok to say here.

We'll also be working through our account rep however since this is a hardware purchase decision I'm not sure how much guidance to expect there at this point.

Thanks for any information, direction, resources or shared experiences!

Accepted Solutions (1)

Accepted Solutions (1)

Former Member
0 Kudos

The best thing you can do here is consider Lumira server is just a thin overhead to provides access into HANA. It's not relevant to the sizing discussion - focus on HANA instead.

For HANA you have two numbers to process - 3bn scans/sec/core and 25m aggregations/sec/core for Ivy Bridge (2/16 for Westmere).

So let's say we have a 2S/30c/512GB Ivy Bridge system. This will give you 750m aggs/sec/core. So suppose you have a 750m row result set and you aggregate the whole lot, you can do this once per second (or 3600 queries/h).

If you have a 750m row result set and only aggregate half of it, you could reasonably expect 7200 queries/h. If you are scanning huge amounts but not aggregating, then you need to focus on the result set.

It's worth noting if you have big data, you need a decent amount of cores to get awesome performance. I have been testing 5bn rows on 40 cores of Westmere and if you aggregate all the data in one query, performance suffers. Upgrading to 160 cores fixed that.

It's also worth noting that if you have very complex queries and joins (unlikely if you are using Lumira Server as it is well normalized) them these numbers can drop.

On average size, my advice is to get at last 30-60 cores, you need this many to experience how amazing HANA is. On compression, it varies. Against CSV flat files, we usually see 5-10:1.

On hardware prices, they didn't fall so much, but make sure your procurement department negotiates.

Hope this helps.

mike_howles4
Active Contributor
0 Kudos

John, thanks for the detailed information.  One question for sake of clarification on this comment I have:

On average size, my advice is to get at last 30-60 cores, you need this many to experience how amazing HANA is. On compression, it varies. Against CSV flat files, we usually see 5-10:1.


We are only licensed to use HANA XS for Lumira Server (LREA limited runtime whatever).  Was your 30-60 core recommendation under the assumption we'd be using HANA for something in addition to Lumira Server, or do we truly need that type of power for just this limited application of HANA?


Thanks again for the great information.  This will help for other (separate) HANA tracks that are moving at a different pace than this Lumira Server track which I see as simpler in scope.

Former Member
0 Kudos

It really depends what you have under Lumira Server. Since you have the power of HANA you have the ability to process large volumes of data.

However you're right, you could use this for a more modest use case. The smallest HANA systems are 2CPU / 30 core / 128GB though unless you use HANA One.

Answers (1)

Answers (1)

Former Member
0 Kudos

Hi Michael,

Answers to scalability questions definitely fall into the realm of 'it depends'. However, I can share some recent testing results from the Lumira Product Group. This was using a 128 GB HANA machine with 16 cores. I.e., not a large machine. It was running HANA revision 82 and Lumira Server 1.19.

Using a 3GB 200 million row data set, we were able to support 200 active concurrent users with a think time of 20 seconds. With a think time of 10 seconds, we were able to support 80 active concurrent users.

Again, this totally depends on the data set and the analytics you perform and of course the hardware you run it on. Also, we're expecting performance improvements with the next release of HANA (rev 83) and Lumira Server.

You should start to see some of these benchmark results become part of the Lumira Sizing Guide, available on www.sap.com/bisizing

Ian Treleaven

Senior Product Owner, Performance and Globalization

Lumira Product Team

mike_howles4
Active Contributor
0 Kudos

Thanks for this information, Ian.  Judging by your response and John's below, it looks like I have quite a spectrum of sizes to narrow down to.  This will help.