Thin Pools, why do they need to be rebalanced after a while?

Question

I have VMAXs in production with thin provisionning for a few month now, and I note a stange behaviour.

In most of our thin pools, the TDAT's does not appear to be all filled at the same level. The skew can go from one TDAT full at 56%, to another TDAT in the same pool at 96%.

I wonder how this could happen, as nothing has been changed in the pools constitution, no TDAT were added or removed, only TDEVs were bound or removed from the pools.

A corrective action is probably do initiate a balancing of the pools, but my questions:

Is there a parameter that would automatically keep the TDAT well balanced?

Is it a recommendable preventive action to start balancing of the pools on a regular basis?

Any suggestion is welcome.

sauravrohilla · Accepted Answer

Alain, the default value is always the recommmended value. variance of 1% is the default and recommended value but in your case the variance was too high. During rebalancing if a tdat is busy handling active IO, vmax skips that tdat and moves on to the other tdat, this causes a slight variance. I believe you must check the audit logs to find out if the tdats were added to the pool?

regards,

Saurabh

sauravrohilla · Answer

In my understanding, when a tdev is bind to a pool and is being written by the host, the writes are spread evenly across all the tdats in the pool. I am not aware of any situation apart from addition or deletion of tdats where one would see this much of variance in allocation on tdats.

I am sure John will have a better explanation

regards,

Saurabh Rohilla

SanjeevMalhotra · Answer

Automated bool rebalancing allows the user to run a balancing operation that will redistribute data across the enabled Data Devices in the Thin Pool. Because the thin extents are allocated from the Thin Pool in a round-robin fashion, the rebalancing mechanish will be used primarily when adding Data Devices to increase Thin Pool capacity.

The symmetrix will move thin deviec extents from the Data Devices with the highest used capacity to those with the lowest until the pool is balanced.

The cause for this vaiance in your case can be due to the Rebalancec Variance figure set to a higher value such as 15% or 20%. Due to this high value while writing extents to tdevs enginuity will write 15% or 20% plus or minus and in such a case you may see the utilization difference of 30% to 40% as the case may be. There could be other internal reasons also maybe which i am not aware.

You can check this value by running the command:-
symcfg show -pool TEST -thin -detail -sid XXX

You can rebalance using the following command:-
symconfigure -sid XXX -cmd "start balancing on pool TEST type=thin;" commit

Please mark this as answered if it answers your query.

Alain3 · Answer

The rebalance variance is set to 1%. I believe this the default when you create a pool.

I did the start balancing, and it does it.

Coul it be that the 1% rebalance variance is too small? Is there a recommended value to set it?

Zikas · Answer

Hello Alalin,

if i am correct, using the recommended Rebalance Variance then you will see an aberration on 1% or 2% on all TDAT devices.

I mean that if one TDAT is 70% full, another can be full at 71%. I am talking about those kind of aberration if you have the Rebalance Variance at 1

emcmagic · Answer

In my case, the Rebalance Variance=1%, but TDAT's are in 67-70% full, there are about 4 variance, should I run the rebalance now?

Other than when new disks being added, are there any situations I need to run rebalance manually?

emcmagic · Answer

Thanks, Quincy56!

Quincy561 · Answer

emcmagic, That is only 3% difference between the most and least used TDAT, I wouldn't bother rebalancing until you saw something like 10% difference.

VMAX

Thin Pools, why do they need to be rebalanced after a while?

Was this post helpful?