Thanks to everyone who joined in our Q&A on IT infrastructure for SAP HANA with Ryan D’Costa, SAP HANA strategist and solution architect at Fujitsu, and Scott McCullough, SAP solutions architect at NetApp.
Review all the questions and discussion in the chat replay here and in the edited transcript, below.
SAPinsider: Looking forward to today's Q&A! You can post your question at any time, then Ryan D'Costa of Fujitsu and Scott McCullough of NetApp will be here, in the chat, today at 11amET/8am PT to take your infrastructure questions around SAP HANA. Today's chat is moderated by HANA 2014 conference producer Bridget Kotelly.
Bridget Kotelly: Welcome to today’s live Q&A! Thanks to everyone for joining us.
Today’s online chat is on the topic of SAP HANA infrastructure – how can HANA impact your discussions around data recovery, servers, sizing, appliances, and the implications for your current hardware and infrastructure? Today we’re joined by Ryan D’Costa and Scott McCullough for their experience and insights on planning for SAP HANA. Ryan is an SAP HANA strategist and solution architect at Fujitsu and Scott is SAP solutions architect at NetApp.
Ryan, Scott, thank you for joining us and taking questions here for the hour!
Scott McCullough: Hello, thank you for having us.
Ryan D’Costa: We are happy to be here - thank you for having us.
Comment From Guest: When does it make sense to have an E-Series based HANA environment, when FAS?
Scott McCullough: Great question... It all depends. E-Series is speed-n-feed without the Data ONTAP abilities. FAS provides the built-in snapshot functionality for HANA. E-Series right now has a higher node density but the new FAS 8000 series also has a higher node count. It comes down to if you need
Comment From Guest: How do I share the binaries for HANA in case of E-series/FCP? Do I need a FAS system for that purpose?
Scott McCullough: That would be one way. You can leverage an existing NFS server, or as you stated use a FAS system for the binaries. If you used FAS, you could also leverage an NFS mount for HANA backup and then use NetApp SnapMirror to move that backup to another location.
Comment From Tribikram Acharya: HANA on ECC 6.0 or HANA on BW - which will be better?
Ryan D’Costa: ECC on HANA and BW on HANA are both viable options to address specific customer business cases. Simply put, if you have significant reporting and analytic issues, BW becomes a better bet. If you have process issues within ECC (like MRP), then ECC on HANA becomes a better business case.
Scott McCullough: This is like an apples vs oranges question. ECC really isn’t meant to do BW type of operations. That being said, I’ve seen shops write custom Z reports to give ECC BW-like abilities (not saying that is the right approach). To have full-functioning BW abilities, I would suggest BW on HANA.
Comment From Guest: What is the greatest advantage that you saw, Scott, in implementing SAP HANA?
Scott McCullough: I've been doing SAP since the mid 90's and we had Z reports that took 12-14 hours to run. I had the opportunity to work an online auction site and those types of reports (similar to my past) were running in under a few minutes. Pretty big "wow" factor, especially given that SAP really hasn't changed much over the years. This is one of those big changes and it's pretty cool being a part of it. Business use cases in this area are a huge advantage.
Comment From Guest: With a TDI infrastructure, do we have to re-run the SAP HW test tool when we add or remove HW and/or HANA systems?
Scott McCullough: Yes, this would be considered a change to the landscape and you would need to re-run the HW test tool to ensure the KPIs are still being met. Adding additional nodes may surpass the controller’s ability to hit the KPIs…
Bridget Kotelly: Scott, can you just define quickly what TDI is for those who don't know?
Scott McCullough: TDI versus Appliance model. How long do we have? :) The appliance model is pretty clear. I say it's a black box that comes pre-bundled and can be running very quickly in a data center.
TDI allows the customer to choose a certified server vendor matched with a certified storage vendor. If the storage and server are on the PAM list, you can use this combo for HANA. The support scheme is similar to an existing SAP landscape: You call NetApp for storage. Fujitsu for servers. SAP for HANA. In the appliance model the appliance vendor (Fujitsu) would get all the calls regardless of component.
Pros and cons to each model. I will say TDI allows more options while slightly increasing the time to setup compared to appliance.
Comment From Guest: For those of us that are not using Fujitsu and NetApp, can you elaborate and/or explain the acronyms?
Scott McCullough: NetApp terms:
E-Series: SAN attached storage for HANA.
FAS: HANA storage leveraging NFS for the persistence layer.
ONTAP: NetApp storage operating system providing snapshots, replication, cloning abilities
Comment From Varma: Can you tell me the configuration of HANA? What is hard disk size, ram size, data volume size, log volume size? How many CPUs required? What processors are used?
Ryan D’Costa: Please refer to this for certified SAP HANA appliances – It also gives you information on data volumes and number and type of CPUs used. For exact configurations, we would need more detailed sizing.
Comment From Ken: Describe disaster recovery for HANA in a VMWare environment.
Scott McCullough: This TR explains the process for recovering a HANA system using the NetApp SnapCreator HANA plug-in. This would get the database back to a PIT. VM or physical environment would be very similar. The OS either on VMDK or physical box shouldn’t have to be restored like the database, similar to a spinning disk database. You would typically only restore the DB, as the OS should be fine.
Comment From Guest: What do you find are the most common SAP HANA use cases in the market today?
Scott McCullough: Good question. Last year I saw more sidecar implementations. Easy to implement. Very low risk. Over the past quarter I'm seeing many more BW on HANA migrations taking place. Very few Suite on HANA so far...but that's understandable.
Comment From Guest: What can I virtualize on SAP HANA?
Ryan D’Costa: Today you can virtualize non production single node HANA environments. Virtualization for multi-nodes and production environments has not yet been released. Virtualized production HANA environments will be released soon and will be announced by SAP and VMware. Please look here for more information.
Comment From Phil C. Any watch-outs on BW on HANA projects?
Scott McCullough: Treat it as an OS/DB migration (if you already have an existing BW instance, say, on Oracle...). If it's a new implementation (somewhat easier), OSS is your friend. Also you would want to decide between TDI and appliance because, depending on your choice, this could impact timelines.
Ryan D’Costa: If your BW environment is fairly large (6 TB and above), I would add more time toward the migration effort - also BI IP environments could add a level of complexity. Please read all relevant OSS notes in the planning phase before the implementation.
Comment From Guest: What are the advantages of running a business warehouse on SAP HANA on 512GB nodes or larger 2TB nodes?
Scott McCullough: If your system can fit into a 2TB HANA node then you could use scale-up. That would cut down costs. My personal opinion is - regardless of whether the data footprint can fit into a scale-up model - the scale-out model provides much more abilities (granted at an added cost).
Comment From Mike: What is the optimal solution for non-production HANA landscape where the non-prod system is a duplicate of production? The requirements are 350 GB Master Node and 880 GB across the slave nodes. Is a Multiple SID on a single HANA Database solution an alternative?
Ryan D’Costa: Yes, that is one possible scenario; another better one from a management standpoint would be to consider a virtualized implementation – utilizing a single individual SID multiplied across a fully virtualized infrastructure running on the non-prod instance. That way it could, at will, in a disaster situation, be quiesced much more quickly with much less operational impact, while simultaneously enabling a more rapid bring-up of the production instance
Comment From Guest: Since HANA tends to be expensive, is it possible to reduce the size of the non-production systems, with only a subset of production data?
Scott McCullough: Absolutely. Many shops I've worked with have a single node system for DEV/QA. For production they have a much larger data set. Just like traditional spinning disk database landscapes.
Comment From Jake: Is there a performance difference when configuring NFS vs. block in a TDI configuration?
Scott McCullough: Excellent question. The way SAP makes all their partners validate storage (and servers) is via KPIs. Regardless of storage vendor or protocol, the storage must meet these KPIs. So in the end, whatever you choose to deploy, HANA will run as intended because of these tests.
And my 2 cents....the tests are extremely high. The fastest performs a ton of I/O and every vendor must meet these KPIs in order to be certified to run HANA. Performance being the same, there are pros/cons with each protocol.
Comment From Jake: What is your take on DR? Using storage based tools or HANA sync functionality?
Scott McCullough: Pros/cons again....
With HANA-based replication you need warm servers at the DR site. You can use these for QA, but it requires running multi SID on your HANA SUSE box. That makes it more complex.
Using storage-based snapshots in conjunction with the HANA Studio provides a less complex SUSE OS, since you wouldn't run multi SID on the DR site. Another way to look at it, NetApp has been doing storage based replication since the mid 90's. HANA has been doing replication for a little over 18 months. Pros and cons....
Comment From Varma: What is the file system used in HANA?
Ryan D’Costa: The BtrFS (Better File System) is recommended to be used for the root file system & EXT (or any other Linux OS FS) for all other file systems.
Comment From Meher: What are the major risks to migrate to ERP on HANA?
Scott McCullough: Great question. This goes back to my roots being responsible for ECC for a global manufacturing company. SAP goes down...plants stop. HR doesn't do payroll. People don't ship product. You get my point. ERP on HANA is new. ERP on Oracle/DB2/SQL (even MAXDB) has been around for years and years... I think HANA is great. I've personally seen what it can do for BW and analytics. It is truly incredible. That being said, SPS7 is the first release with log reply that became available late 2013. My personal opinion I would like it to bake a bit more before I would leverage HANA for ECC. BW or analytics are important for shops, but they do not typically run the business like ECC.
Comment From Guest: My first question is about the sizing approach for ERP suite on HANA or HANA sidecar. The existing SAP notes use the current root data size in the existing classic DB to estimate the required DB space on HANA, considering the compression rate and so on. There is no reference to write/update workload. Does this mean the number of concurrent users and write/insert workload does not impact the sizing?
Ryan D’Costa: No. The concurrent user session count & insert (not so much write because storage has already been used) DO affect sizing of memory and subsequent storage.
This should be viewed with the usual capacity planning and performance management tools, to maintain control over the HANA infrastructure and permit and enable proactive storage and memory upgrades, so as to not degrade performance unnecessarily.
Comment From Phil C.: Working with SAP on a BW on HANA project, are there any particular areas that we need to be aware of as we implement the project?
Scott McCullough: If this is a migration from an existing BW landscape, treat it as an OS/DB migration and plan accordingly. Little easier if it's a greenfield implementation. At a high level, HANA is just another database (albeit in-memory and very fast).
Side note....HANA is the most exciting thing (at least for me) in the past decade working with SAP.
Comment From Guest: In addition to ECC/BW/BPC on HANA, are there any other SAP applications that run on HANA?
Ryan D’Costa: CRM, PLM and SCM. Please check OSS note for limitations (SRM planned for future).
Comment From Guest: Is there a sidecar available for just MRP?
Scott McCullough: You can "HANA"fy an existing spinning disk database. Check out Amazon: SAP HANA Essentials (free download). The book explains how you can make a call into HANA for specific tables (like MRP) to vastly increase performance w/o doing a migration. You update your ABAP to make a query to HANA instead of the underlying DB. The book is a great fast read on the entire HANA project.
Comment From Guest: What's the migration strategy/best practice for migrating BW/ECC from traditional RDBMS to HANA?
Ryan D’Costa: You will have to check on your current version and patch level based on the consideration if you have to upgrade before the migration or run a Unicode conversion - Please check the DMO option for SUM - this can be good option if you meet the pre requisites
Comment From Dennis: Does HANA actually run in-database statistics or just accelerate reports creation in BW?
Ryan D’Costa: HANA does both. HANA will accelerate all queries placed against it (the in-memory) database, however, when BW is being used, it will massively enhance the speed of BW report generation – usually cutting the SAP report generation from several hours to just a few minutes, by using its parallelization abilities.
Please also note that if you optimize DSO objects you will also see significant improvement in data loading too.
Comment From Ken: Describe various DR scenarios with Hana. Can Hana be used with Zerto replication to a remote site?
Scott McCullough: I've not yet seen Zerto in any HANA deployments so far. That's not saying it couldn't work though. With NetApp appliance model or TDI model I typically see the NetApp HANA plug-in providing HANA based snapshots in conjunction with storage based snapshots. This TR has a lot more info.
Comment From Varma: What are the prerequisites for adding a node - explain that process?
Ryan D’Costa: This will usually be undertaken by the vendor. Fujitsu has a professional services team dedicated for these types of solutions. Usually the customer is not responsible for adding nodes to an existing cluster, as it could involve rewiring of the networks.
Comment From Meher: What are the parameters that my impact the required HANA DB? Root data size, number of users, insert/update workload? And how to integrate these parameters for HANA DB sizing?
Ryan D’Costa: The SAP HANA Server Installation Guide can be found here.
Provides most of the guidance necessary to define the initial layout of an SAP HANA system, however, best practices suggest you continually monitor these file systems for rates-of-change data so as to adjust for customer specific, unpredictable, changes based upon their loadings.
Comment From Guest: Any suggestions on combining a HANA sidecar implementation with an existing cDOT environment?
Scott McCullough: Sounds like you have an appliance for the sidecar and you want to leverage an existing cDOT environment. The appliance model can't be tied into another storage controller outside of the appliance model. You could leverage the cDOT environment (at long as the controllers are TDI certified ) for a different HANA environment, though.
Comment From Guest: What is the most common infrastructure setup for an SME (i.e. $200M company) running SAP Suite on HANA?
Ryan D’Costa: Most SMEs combine DEV and QA on the same appliance (MCOS or VM) and have a separate Production environment. However, if you don’t have an IT staff to maintain HANA systems, a Fujitsu cloud solution could also be a viable option.
Comment From Guest: What is a realistic entry level price point for hardware that supports the use of SAP HANA?
Ryan D’Costa: Pricing will be based on the size of your environment (Dev, QA and Production). Fujitsu/NetApp will be happy to provide you pricing for your specific environment.
Comment From Guest: Do you need to clean up your data prior to go-live for the memory database? (Is the cost by size?)
Ryan D’Costa: For BW it’s very important to clean up PSA and logs - you can save a significant amount that could have an impact on your sizing and, ultimately, cost.
Bridget Kotelly: Thanks again, Ryan and Scott, for joining us and taking some time out for the Q&A today!
Scott McCullough: Thank you....my typing skills have increased ten-fold after this session! If you need any more info regarding solutions for HANA or SAP you can find them here.
Ryan D’Costa: It was a pleasure being here - Feel free to contact me directly at firstname.lastname@example.org if you need any additional information. Thank you
SAPinsider: Again a big thank you to our panelists, Fujitsu's Ryan D'Costa and NetApp's Scott McCullough. You can learn more about their collaboration here.
Some great questions and discussion today - thanks to everyone who joined us!