How long does a HANA instance startup take?

[Former Member]

In a previous thread Lars Breddemann wrote:

Data stored in row-store tables will be loaded into the main memory immediately on instance startup.
Data stored in column-store tables will be loaded upon first access, except when it has been specified that the table should also be loaded during startup.

Even if the persistent storage is located on SSDs, doesn't that take a really long time to start a large HANA database?

Is there already some rule of thumb about how long it takes in real life to start a large HANA database?

Or can you simply say that HANA doesn't impose a bottleneck (or other significant delays) but startup primarily depends on reading the data from persistent storage and thats it.

In my imagination, many HANA customers would want to load almost eveything from column storage during instance startup. So imagine a 1 TB HANA database reading 1 TB from SSDs and also creating indexes for column storage during instance startup. What is the order of magnitude for instance startup: Seconds, minutes or hours?

Regards,

Mark

[Former Member · Accepted solution]

Not surprisingly, real life experience with HANA is still rare. Let me do a best-case calculation in order to get a lower limit for the HANA DB startup time:

Example: HP ProLiant DL980G7 with 4x 320GB FIO SSD storage and 1 TB RAM.

The SSDs can perform 800 MB/s each, combined 3200 MB/s.

Now it takes at least 330 seconds to read 1 TB of data from the 4 SSDs in parallel.

Most probably there will be some more overhead involved by HANA e.g. by calculating indexes etc.

[Vitaliy-R · Accepted solution]

Mark, if I am not mistaken by default tables in HDB are marked as "do not load into memory" or so, i.e. on start-up only tables explicitly marked to be loaded into RAM will be taken. All the rest will be read from storage on request. I need to double check that. -Vitaliy

[Former Member · Accepted solution]

I got some feedback on HANA startup times from [Vijay's blog|http://www.sdn.sap.com/irj/scn/weblogs?blog=/pub/wlg/26606] [original link is broken] [original link is broken] [original link is broken];:

2011-09-27 07:12:58 Vijay Vijayasankar Business Card [Reply]
Hi Mark
Yes - November 7th is when it goes into ramp.
From what we saw in POCs, it does not take a long time to start like with traditional DB. These are all things SAP should benchmark and publish, and I am told it will be done soon
Cheers
Vijay 

Looks like I was mistaken, thanks also to Vitaliy for the clarification. So the SAP HANA DB startup is indeed fast, similar to usual RDBMS. Maybe the first user who selects data from a columnar table will notice some delay, we'll see. Other RDBMS need to initially load table contents after system startup into the buffer cache as well, so maybe this is no big deal.

[former_member93896 · Accepted solution]

Hello Mark,

here's some more background: Tables are either in row or in column store. Row store tables are loaded into memory during startup. Column store tables are set as a default to be loaded on demand i.e. the system loads columns when necessary. You can change the preload behaviour for column tables using ALTER TABLE (see http://service.sap.com/hana > SQL Reference).

Since most data will be stored in column tables, actual startup time is indeed pretty quick. If you want to avoid a performance hit on first data selections, you could create a simple SQL script that executed a LOAD statement your most critical tables and run it once after start up and when it fits best to your requirements.

Regards,

Marc

SAP Customer Solution Adoption (CSA)

[Former Member]

This message was moderated.

[davidebruno]

This message was moderated.