What is a "Shared Storage" in a HANA system

[BenedictV]

Hi,

I was going through a document on high-availability of HANA systems and I came across the following,

My question is, what does the shared storage mean here? Is it an external back-up location that is shared across the servers or is it the disk storage of the servers that is being shared?

Thanks,

Benedict

[former_member184871 · Accepted solution]

Hi Vasanth,

As far as I know and understood The Concept is :

It depends on HANA Landscape, from your picture - I can assume that - its a shared disk storage which is connected with different nodes of HANA within the same HANA site.

SPS05 - Onwards

1. Storage based mirroring of SAP HANA disk areas controlled by storage technology
2. WARM standby: DATA and LOG content is continuously transferred to secondary site under control of SAP HANA database
3. HOT Standby: DATA content is only initially transferred to secondary site, afterwards continuous LOG transfer and LOG replay on secondary site

For more detail you can also refer HA on SPS06  : Introduction to High Availability for SAP HANA | SAP HANA

The SAP HANA database system supports high availability with a failover scenarios for hosts. It is possible to  have  standby  hosts  that  can  take  over  if  an  active  host  should  fail  –  for example  because  of  hardware failure.  The  standby  hosts  work in “cold” standby mode. This means that they just register with the Master Name Server and wait.

As long as they are in standby mode the Index Servers on these hosts do not contain any data and do not receive any requests.  When an active host fails, this is detected by the Master Name Server which selects a standby host to take over. The Index Server on the selected standby host takes over the database volumes of the failed Index Server and loads the data into memory by executing the restart sequence.

Assigning the volumes of the  failed  host  to  the  standby  host  requires  having  the  database volumes  on  a  shared  storage  server. Therefore  this  high  availability  scenario  is  called  cold

standby  with shared  storage.

For large amount of data, loading the data into memory during failover may take considerable time. In “hot” standby mode, the standby server would be kept close to the state of the original server to allow a very fast takeover.  This  would  require  replicating  all  changes  to  the  standby  server during  normal  operation,  for example  by  replicating  the  transaction  logs  and  replaying  them on  the  standby  server. But  what  happens  if  the  host  with  the  Name  Server  fails?  To  ensure high  availability  of  the  Master  Name Server, up to two Slave Name Servers can act as backup Master Name Servers. If the current Master Name Server  fails,  the  backup  Master  Name  Server

takes  over.  If  two  backup  Master  Name  Servers  are configured, they negotiate and one of them takes over the role of the Master Name Server. The new master then performs a restart from the persisted data and transaction logs of the old master. This is similar to the process described above for the Index Servers. You may wonder why the new Master Name Server needs to restart??  As  a former  slave  server  it  received  all  data  already  via  replication.  The  reason  is  transactional consistency: replication is not guaranteed to be immediately consistent. The most recent changes might be missing  in  the  slave  name  server.  Only  by  restarting  from  the  persisted  data  it  can be  assured  that  all committed changes are restored.

Regards

Kumar

[BenedictV]

Thank you all for your inputs. Now I have a better understanding about the disk storage in HANA

[karthikeyan_natarajan4]

Hi Benedict,

In HA system, you can have one single storage for many servers, where all the data and log volumes will be created separately. For example, for server 1, there will be a separate data and log volumes for each service (index server, name server, statistics server and XS engine). If the host is configured as standby it won't have the data and log volumes, In case of failover, name server of active server will the map the data and log volumes of failed server to standby server. That's the reason they have directory structure as mnt00001, mnt00002 for each server and sub-directory as hdb00001, hdb00002 for each volumes (which basically represents volumes of each individual service)

Regards

kartik

[nicholas_chang]

Hi,

Adhere to the above points, we've setup HA environment in VM as playground, and our shared storage is on NFS. Basically, all nodes within the HA group are accessing the same /data and /log file system.

Cheers,

[vivekbhoj]

Hi Benedict,

As explained by Mayuresh above, it depends on HANA Landscape.

Different Hardware partners have implemented High Availabality differently as IBM uses GPFS(General Parallel File System) -  a shared disk clustered file system developed by IBM itself while HP, Fujistsu use different implementation.

You can also check the following documents:

http://www.saphana.com/docs/DOC-2935

The below whitepaper is from Oracle on their Analysis of HANA High Availabilty:

http://www.oracle.com/technetwork/database/availability/sap-hana-ha-analysis-cwp-1959003.pdf

Regards,

Vivek