Presently Kolkata Tier-2 @ ALICE contain one eos named as ALICE::Kolkata::EOS with EOS instance name is “kolkataalice”, configured as RAIN-6 configuration. EOS version is EOS 0.3.268 (CERN). This was configured during beginning of 2016 and its full now.
Configuration is as follow:-
There are three eos fst disk server, named as eos01.tier2-kol.res.in, eos02.tier2-kol.res.in and eos03.tier2-kol.res.in, and one MGM server i.e. eos.tier2-kol.res.in. Each fst has 12*4TB HDD, total 48TB in each server. and its total storage size is 129TB after File System.
Now to extend storage capacity under Kolkata Tier-2 we have procured 7 * 2U DELL PowerEdge R730XD servers, each comprising 2E5-2630 processor, 1610TB NLSAS disks for data, 2480GB SSD for OS and 128GB RAM. Each server contain 160TB and total 7160 = 1120TB Raw.
What is the best way to configure this, with existing EOS:-
a) Can we add these new storage servers as only FSTs in existing EOS MGM or in existing EOS instance?
b) Can we add 7 similar FST servers under EOS RAIN-6 configuration or we have to use 6 similar servers only?
Hi Vikas,
if you want to continue to use the RAIN layout, I would add the 7 new servers into a new space (not the default one) on the same MGM. So you will have the default space with 129 TB and the new space with 1.1TB raw capacity.
You could profit from the update to put the latest version of EOS and migrate to the new namespace backend. Then you could use your three old machines as Quarkdb hosts and MGM failover nodes. You get a better high-availability in this setup with 3 fold redundancy. The three old machines still can serve the old files, but you reconfigure the ALICE grid directories to write new files to the new created space with the 1PB space.
You can also do this is two steps, first just upgrading to CITRINE and create the new space for the new capacity. Then in a second step you do the a little bit more complicated upgrade to use the new namespace backend with the high-availability implementation.
if you want to continue to use the RAIN layout, I would add the 7 new servers into a new space (not the default one) on the same MGM.
Yes I want to use RAIN layout because as per my understanding RAIN is great substitute of high end SAN based storage with really low budget.
If we implement 2 instance in the same MGM, then hardware wise what is configuration of MGM. Probably we need to change old MGM server also, but we do not have any new spare server. In that case should we use one of the 7 server as MGM together with FST?
You could profit from the update to put the latest version of EOS and migrate to the new namespace backend.
Latest version of EOS is CITRINE and it support namespace backend? Which you suggested in step 1.
migrate to the new namespace backend. Then you could use your three old machines as Quarkdb hosts and MGM failover nodes.
Could not get it clearly? What is use of Namespace backend, Quarkdb hosts? kindly explain in brief and provide more PPT or paper for the same.
MGM Failover node means MGM slave? or is it something else.
You get a better high-availability in this setup with 3 fold redundancy.
How it is 3 fold redundancy. One is at RAIN-6 level, which are other two?
The three old machines still can serve the old files
Means the old eos instance is in read only mode, right?
Old 3 machines are almost full, then how we will use these as namespace backend? What is minimum criteria for this?