Object store and Shadow Memory

[Former Member]

Hi HANA experts,

When i was reading about Object Store and Shadow Memory concepts in HANA, I could not get much information about these two and not quite understood it also. Kindly provide some help on these 2 concepts.

Thanks in advance.

Regards,

Vivek

[Former Member · Accepted solution]

Hi Vivek,

A fairly open ended question but ill do my best to try and answer it.

The concepts you mention above relate to the Persistence Layer in HANA. As you are aware HANA is quick because unlike traditional databases it stores data in RAM (memory layer). This is great, except in the event of a power outtage or other event. if this happened with no data saved to the disk then all changed, visible data would be be forever gone.Luckly HANA has a layer called the persistency layer which handels page management a logging (redo and undo logs) and permanently stores data in disk storage outside of RAM. Hence the relevance of the name persistence. The disk has seperate storage volumes for data and logs. 

Within the in-memory computing engine HANA makes use of both column and row based storage for data. There is a row storage (EG like a traditional data base store but in RAM) and a column storage (better performance, compression and read abilities with processing). These 2 stores make use of the persistence layer in order to ensure all memory based data is saved to disk.

Effectively HANA regularly saves all data in memory every few minutes (Can be configured to desired time 5 min, 10 min, etc). The persistence layer consists of a logger (information about changes) and page management. The persistence layer stores data in disk volumes that are organised in pages. Data pages and before images (undo log pages) are written on data volumes on the disk. After images (redo log pages) are written to the log volumes on the disk.

Data is written to the disk either when:

1) transaction ends and the redo logs (after image) are written to the log volumes.

2) savepoint is performed and changed data itself is written to the disk,

Shadow memory concept. This relates to the savepoint data write to the disk. EG all data which has been changed since the last savepoint will be marked and written to the disk (delta mechanism). The data belonging to the last savepoint will not be overwritten. EG Shadow page concept.

Hope this provides soem clarity.

Kind regards,

Danielle

[Former Member · Accepted solution]

Hi Danielle,

Thanks for the information. It was quite helpful as a start-up. But had a few other doubts while reading through.

Actually when there is a power outage, all the committed transactions are saved in the Redo logs. That's what I read in a book. But you have mentioned that all data will be marked since the last savepoint and written to the disk.

Here What is the difference between logs and disk.

Regards,

Vivek

[Former Member · Accepted solution]

Logs are transaction logs and always stored synchronously (with the transaction commit) on disk. With disk the actual database data is meant (say table data), often called the data files. The data files are updated asynchronously.

This is just like almost any database guarantees consistency.

Cheers Michael