Start a Conversation

This post is more than 5 years old

Solved!

Go to Solution

2089

October 6th, 2014 07:00

How does FAST move the extent?

Hello Experts!

Could you please tell me how it can be according next output if we have the FAST policies like this:

Policy A:    10% EDF - 10 % FC - 80% SATA

Policy B:     15% EDF - 15 % FC - 70% SATA

and output:

Screenshot002.jpg

the device 09c1 and 061e are bound to SATA pool and as you can see the extents are allocated on SATA and EDF only?

How is it possible? We have strong rules for FAST and as I understand all extent should be allocated according these rules:

(for Policy A) - 10% extents should follow to EDF, 10% to FC and 80% to SATA respectively. On the screen shot I don't see allocation for FC pool at all. Could you please explain?

And one more question regarding pools and FAST.

For instance we have 3 pools - EDF, FC and SATA. and free space on them EDF - 1 gb, FC - 5 gb and SATA - 100gb.

and the same FAST policy - 10% EDF - 10% FC and 80% SATA.

I bind new volume 100gb to FC pool (Max. Subscription limit allows me to do that and don't think about PRC for this example) and host does write to volume and within 10 min all 100 GBs have been written to volume. Of course almost all of extents will be written to SATA pool (am I right?) and then how the FAST will transfer extents if we have so strong policy? it can allocate 10 gb on EDF and 10 gb on FC pools, there are not enough space. which is the behavior of  FAST in this situation?

Thank you!

Message was edited by: evgenym

51 Posts

January 21st, 2015 08:00

Summary:

FAST works on the storage group level, it means all your volumes can be located in a different pool in spite of their size or amount. For instance if FAST mechanism decided according skew to move 50% of extents of one volume to EFD and another part (50%) of that volume on SATA, it will work. So FAST looks on the storage groups and all extents within all volumes like one big entity with a lot of extents. That's why possible to have situation when only volume could be:

- only on one pool (level)

- on 1st and 3rd level (2 and 3, 1 and 2 too). So storage group for him is like one big volume.

In another hand, if space (on pool) is not enough for promotion/demotion of extents the FAST will keep trying to follow policy. In case if it is not able to follow policy (migrate groups of extents) it will stay as is (keep extents on old places and wait, when space will be enough for promote/demote). Also FAST looks on the response time of pool before moving of extents, for more information please read there: https://community.emc.com/thread/169450?start=15&tstart=0

2.1K Posts

October 15th, 2014 23:00

hi evgenym,

Please consider moving this question as-is (no need to recreate) to the proper forum for maximum visibility.  Questions written to the users' own "Discussions" space don't get the same amount of attention and questions can go unanswered for a long time.

You can do so by selecting "Move" under ACTIONS along the upper-right.  Then search for and select: "Symmetrix Support Forum" which would be the most relevant for this question.

51 Posts

October 16th, 2014 02:00

Thank you! I moved it. Should I do it for each topic? Have I created in wonrg flow, right?

51 Posts

October 16th, 2014 04:00

Ok, this is clear for me I though the same but was not sure, so VMAX can promote or demote the extents if pool has no space. But what about first part of question, I'm totally confused. As I remember from VMAX course if array has strong rule (the sum for all tiers in policy equal 100% in my example 10-10-80) it will keep the written extents according this one. For instance we have 100gb allocated device and written 50 gb, so according policy vmax will write 5 gb on EDF, 5 gb on FC and the rest on SATA 40gb (80% of 50 gb). But why I dont see any written extents on FC pool, I have no idea

859 Posts

October 16th, 2014 04:00

For the first part of questions, its really strange for the extent to be promoted directly to EFD. It could be possible that the 10% allocation for FC tier is already full for the policy to which this device is associated or FC pool has already hit 90% of its total capacity.

Another possibility is that the extent was active enough to be promoted to EFD directly.

For instance we have 3 pools - EDF, FC and SATA. and free space on them EDF - 1 gb, FC - 5 gb and SATA - 100gb.

and the same FAST policy - 10% EDF - 10% FC and 80% SATA.

I bind new volume 100gb to FC pool (Max. Subscription limit allows me to do that and don't think about PRC for this example) and host does write to volume and within 10 min all 100 GBs have been written to volume. Of course almost all of extents will be written to SATA pool (am I right?)

Right, all new writes will go to bound pool unless you have -vp_allocation_by_fp parameter set.

If tiers are full then promotion cant be performed but demotions continue to happen. VMAX code polls sub-lun statistics from all the devices in SG and generates threshhold per tier. Symm calculates promotion score and demotion score for each extent per tier.
Based on this score and threshold, VMAX enginuity knows which extent to demote and promote. If there is not enough space in the tier to move an extent then the threshold for promotion might increase, thus allowing extent with the highest promotion score to be moved to the efd tier.

feel free to add and correct me if i have missed something.

regards,

Saurabh Rohilla

October 16th, 2014 07:00

Ok, there is no flow which can be defined to move extents from one pool to another pool and also this is not necessary to move extent should into the FC or bia FC to EFD, this is completely depends on IOPS and according to that extent can move directly from SATA to EFD or FC!! Possibility is that the extent was active enough to be promoted to EFD directly.

Please feel free to add and correct me if i am wrong!!

2.1K Posts

October 16th, 2014 19:00

hi EvgenyM,

if you have other threads created in your personal space, please move them too. For new threads, just go to the related space then click new discussion.

51 Posts

October 20th, 2014 08:00

Hello!
Yes, I can agree with you that not needed to keep the tracks on FC pool, but from a course I remember if VMAX see strong rule (sum is 100%) it will place all extents according this policy (for my example 10% should be on EDF, 10% on FC and 80% on SATA) so your answer is not clear for me, I can understand you, but I dont understand the behavior of VMAX in this case, seem you too if you, can you tell me the document which describes that? I couldn't find it on EMC support site.

November 13th, 2014 03:00

Hello!

Sorry but it is not possible, if FAST has strong policy it should follow of this policy

34 Posts

November 13th, 2014 05:00

FAST operates at the SG level not the device level (unless you have an SG with a single dev in it), devices within the SG may have different allocations but as long as the group percentages fit the policy it will be within tolerance.  

51 Posts

November 18th, 2014 04:00

Hi Chris!

Seems I disagree with you, what do you mean when you say FAST operates at the SG level? Actually you just associate SG with proper FAST policy, but FAST mechanism works on sub-lun level with extents. (actually it calculates the performance for each block 64kb each one - this  is minimal entity which VMAX operates (slot, IO rate and so on)) and then migrate the groups of extents (10 items, about 8 Mb). So I would say each volume should follow of this policy, am I right?

or FAST operates only one entity like SG?

51 Posts

November 18th, 2014 04:00

Hi again Chris!

I'm not sure but if FAST operates one entity (SG) as you mentioned, does it mean that VMAX looks on the volumes within in one SG like one wide device with a lot of extents and then migrate them according rate of IO for each one?

so for example that to be clear. We have one SG and three devices within A,B,C, and extents for each one:

A -> A',A'',A'''

B -> B',B'',B'''

C -> C',C'',C'''

Does it mean that FAST sees this only like that:

A',A'',A''',B',B'',B''',C',C'',C'''  and in this case it makes sense to work as I described: VMAX can move extents like this:
SATA: A',A'',B',B'''

FC:  nothing

EDF: A''',B', C',C'',C'''

Am I correct?

Thanks!

859 Posts

November 18th, 2014 18:00

You both are correct. As Chris said, FAST VP works on SG level, yes it does work on SG level to calculate the policy allocation percentage and extent level to measure performance as you pointed out.

So if a policy is created to have 50% allocated to SATA and its applied to a SG. Then the 50% of the total SG capacity (excluding pinned devices) will be moved to SATA not necessarily 50% of each device in that SG.

regards,

Saurabh Rohilla

51 Posts

November 21st, 2014 07:00


Hello!

Please tell me does it mean that if I have strong policy and not enough space for allocation on FC pool for instance, then Enginuity will not follow the rule, am I correct?

859 Posts

November 23rd, 2014 21:00

As I said in my reply on Oct 16, promotion & demotion are based on the threshold scores for each tier. If the tier is Full, there wont be any promotion (even if the policy is strict).

For e.g. You have 1 TB of total SG capacity and your policy says 100% to EFD (total size of EFD tier is 500 GB) and 1% to FC tier (bound pool). What will happen in this case? Will FAST move 1 TB to EFD that has only 500 GB?

regards,

Saurabh Rohilla

No Events found!

Top