Dual purpose HA/DR for large scale-up HANA

[Former Member]

Hi all,

We have a potential project that is looking like it will require a 12TB system. It's well known to use a QAS system as the HA or DR target to save on having to purchase additional nodes. I understand the architecture will depend on the customers BC/DR requirements. I  have numbers on the additional resources the replication process would necessitate be available on the target system. I also realize, I should talk with the vendor, etc...... 

🙂

Of course, we would rather not have a 12TB dedicated node sitting in a secondary DC that will (hopefully) never be used.

Please share any real world experience implementing a dual purpose DR appliance at a 12TB scale and specifically if there are any considerations that may need to be looked at that smaller scale-ups may not require.

Thanks!

-Jake

[Former Member · Accepted solution]

The short version is that this is a workable solution from a technical perspective. As with other configurations, you will need double the disk - or an equivalent TDI model for disk. When we did testing, we found that the storage KPIs aren't sufficient for real-world performance (yet) and so you'll need a lot more IOPS than SAP recommend to make this work well. This will need some tuning.

The problem I foresee with your scenario is that 12TB appliances can be very slow to start if cold. I've seen ~2h+ start times to load from memory to be completely running. Would that cause you an unacceptable RTO? If so, you could maybe run QA at 8TB and Prod at 4TB (cold). Then if you failover, at least the major tables would be loaded and you could keep RTO acceptable.

I've not tested this in SPS10 and it looks like they have made strides in this area, so this might be better. Either way, depending on DR RTO, you may need a hot DR system. Given the street price of a 12TB box is over $1m, that could be costly.

The only other thing you need to bear in mind is that you need fantastic network between data centers. In certain circumstances, where you failover in the primary, and want to switch back, you might have to do a full sync.

[Former Member · Accepted solution]

Thanks John. Very insightful.

Not many in the world have experience at that scale. Appreciate the feedback.

[Former Member · Accepted solution]

NP. I should mention we implemented this exact scenario with HA/DR and SUSE automated IP-based failover at 6TB scale. It works very well.

The HA is automated and the DR is manual - that's a deliberate choice and most customers won't want to automatically initiate the DR process.

HA takes a littler longer in SPS09 than I would like - around 2 minutes. The enhancement list on SPS10 suggests that this is faster now - they improved the row store takeover mechanism.