This post is more than 5 years old
1 Rookie
•
13 Posts
0
14154
June 22nd, 2011 15:00
How data is written to avamar nodes?
Hello All
I have a 16 node avamar grid in my environment, I would like to understand how data is written to the nodes and to the disks. And how can we recover the data incase of node failure?
Thanks in advance
Kuldeep
No Events found!
rpervan
266 Posts
0
June 26th, 2011 03:00
Kuldeep,
> I have a 16 node avamar grid in my environment, I would like to understand how data is written to the nodes and to the disks.
The "avatar" is the primary process for backups and restores. avtar communicates with the gsan processes on the storage nodes.
During a scheduled backup or restore, the administrator server (mcs) running on the Avamar server generates a work order. mcs then pages the client avagent and avagent picks up the work order.
On the client, avagent starts avtar to begin the backup process. avtar communicates to a gsan process running on one of the Avamar data nodes and sends backup data to the gsan. The gsan process then distributes the data across the available data nodes in a multi-node Avamar server.
> And how can we recover the data incase of node failure?
Each backup has its own root hash linking to the data objects and hashes which comprise the specific backup at the point in time when the backup occurred.
For a restore initiated by Avamar Administrator, MCS contacts avagent on the applicable client. avagent starts avtar and avtar connects to a GSAN on the Avamar server. GSAN processing locates the chunks necessary to fulfill the restore request and sends the chunks to the avtar process running on the client. The chunks are collected in the client’s memory, ordered, uncompressed and written to disk on the client.
Data is stored using a complex hierarchical hashing file system with an indexing structure consisting of the data elements, grouped together by multiple levels of hashes, composites and root hashes. The root hash of each backup links to the data objects and hashes comprising the backup at the point-in-time when the backup occurred.
Composite hashes are stored in separate stripes. Root hashes, as well as information about the origin of the files (client, domain, etc.), are stored in the accounts stripes. On a RAIN system, an additional stripe file contains RAIN parity data. This data is used to reconstruct data for a failed node. These additional stripe files account for the RAIN overhead.
regads,
.r
Avamar Exorcist
462 Posts
0
June 27th, 2011 07:00
Avamar Exorcist
462 Posts
0
June 27th, 2011 07:00
My last reply got garbled so let's try again..
If data node hardware fails --> Replace the affected part or perform a chassis swap using the spare node hardware. Some hardware items are considered user replaceable, others (such as RAM) are not. See the Avamar Datastore customer service guide for more details and consult your EMC Partner or Support if you have questions.
If the hardware of a utility node fails --> The utility node can be rebuilt as described in the following document on Powerlink
Home > Support > Technical Documentation and Advisories > Software ~ A-B ~ Documentation > Avamar > Technical Notes > Avamar Utility Node Replacement Procedure v5.x Products and Earlier Technical Note. Would recommend involving your EMC Partner or Support before carrying out this procedure
If some kind of software fault occurred within the Avamar GSAN (which for some reason cannot be fixed), the system can be rolled back to a previous checkpoint. If the failure affects only one data node then, providing this is a RAIN array, a rebuild of that failed node's data can be initiated using the parity from the other nodes within the grid. This must only be performed by EMC Support.
Kuldeep_Singh1
1 Rookie
•
13 Posts
0
July 6th, 2011 16:00
Thanks Rej and Nicholas for providing detailed information on the process flow.
I understand that it is possible to recover a failed node with the parity at Node level and a failed drive with parity at drive level, but is it possible to recover the grid if there is a failed data node and a failed drive on that data node?
TA
rpervan
266 Posts
0
July 7th, 2011 00:00
Hello K_S,
> but is it possible to recover the grid if there is a failed data node and a failed drive on that data node?
Of course, in that case you just need to replace failed hard drive ... RAID 5 should properly rebuild data structure on failed hard drive . The data should be consistent after RAID rebuilding.
.r
Kuldeep_Singh1
1 Rookie
•
13 Posts
0
July 16th, 2011 10:00
Thanks Rej