HPE Storage Users Group

For myself, with only a few drives in the array I would go with a 3+1, I can't see the additional parity writes introducing substantial overheard. I believe that Markinnz was making the 6+1 recommendation because it meets the requirement to be able to lose a single disk.

I am not privy to your workloads so I think it is difficult to make recommendations based upon the limited information gleaned from the post and because there are so many factors involved.

-Do you have a business reason that is pushing you to AFA?
-If your need is mostly random I/O then AFA may be a the best choice
-If your need is write heavy why not consider an 8440 with AFC?
-The 8440 and 8450 both have the same CPU and memory but the 8440 is more flexible because of AO and AFC.
-You are using Peer Motion, are you doing RCIP or RCFC?
- If RCIP are you using only the 1GB RC ports that come on each controller or have you added additional ports? If you are just using the single 1GB ports then depending on the amount of change you could be introducing a bottleneck, maybe consider purchasing the 10GB card.

I would recommend looking at the 8440 with a solid SSD tier used for AFC and VV's, placing only your I/O heavy VV's on SSD. If you have heavy random read or heavy write that is not accessed frequently then I would avoid AO. No matter what you go with consider using smaller drives and build so that at most it is only 50% populated to allow for rapid expansion.

At the end of the day recommendations differ between admins, not that one is right and the other is wrong but because we have different experiences and needs. One thing I have learned is I NEVER blindly trust solutions provided by a VAR or the MFG sales rep. I have been left with "workable" but sub-par solutions too many times so now I "own" all of my solutions, with that being said they can be fantastic resources to utilize and learn about various options and risks.

_________________
Adam Berus - I.S. Tech Lead - Wolverine Worldwide Inc.
x1-7200, x2-7400, x1-7450, x1-8400, x2-8440
3PAR Technologies: AFC, AO, FilePersona, RemoteCopy (RCIP)

Thank you for your interesting post aberus.

Do you have a business reason that is pushing you to AFA?
Those nodes store our VM's and all our databases (normal VM in LUN's with dedup+compression and database VM's in LUN's with compression only)
We had lefthand arrays before and at the end of the contract for these arrays we decided to go for an AFA because SSD's are more and more cheaper and for us it was the right time to jump in.
With dedup+compression on two 8200 nodes we were able to replace dozens of LeftHand arrays of 15K HDD's.

We hesitate between some AFA vendors but we go for HPE because of the active/active nodes with peer persistence.

-You are using Peer Motion, are you doing RCIP or RCFC?
We are using RCFC and it is working very well.

The funny thing is that we asked for an HPE engineer to come on site to verify our arrays (settings, cpg, VV's ...) before we put them in prod and he did not notice something like this...

And we thought that compression was an ASIC process, no one told us it was a CPU intensive workload which put 8200 arrays on their knees... and at that time it was not documented (i have not checked since...)


If we go for a 3+1 with 8 disks does it meets the requirement to be able to lose a single disk ?
If we would like to extend our array with this setting, how much disks should we buy per node ?

Yeah I've seen a lot of people being pushed 2 node systems and then using heavy IO mixed with all the features plus SSDs and being disappointed in the results. They also like to push the bigger drives but this has a lot of gotchas attached that only come out after it's up and running for a while.

I find there's a growing disparity between the marketing, theoretical, realistic and reality of 3PAR systems. It's still good stuff but you really need to be able to sanity check what's being sold to you as certainly what I've been recommended in resent years wouldn't last the expected lifetime in my experience and would need replacing rather then upgraded.

Initial savings of 10-20% are great in spec tweaks but if it needs replacing in 1-2 years... not so much.

Hi Chris

I would imagine that most all of your dedupe savings are coming from your VM's and probably very little from your databases. Since dedupe is VV based you may want to consider only enabling dedupe on your VMFS VV's and using physical RDM's for your databases.

Dedupe is a function of the ASIC, if I had to guess your high CPU is related to compression and remote copy. I do not use dedupe or compression on my AFA (7450) because it is dedicated to large Oracle databases where I would receive no benefit by utilizing it.

In my experience, VMware is an awesome use case for implementing into a tiered environment leveraging AO for performance because there is a high(er) level of predictability. Where this starts to fall apart is when transactional systems are deployed because it can skew the I/O density results which would be another use cause for leveraging RDM's for those VM's

About the only case I can make against a multi-tiered array (like the 8440) would be skillset and management overhead. If you just want something that just works then AFA is the way to stay.

I'm sure there are some awesome HPE Engineers out there but my experience with them has been less than stellar to the point where I have a waiver in place where I perform all of my own installs and upgrades so all installation services are removed from my purchases. The ones I have worked with seem to function off of a script and have issues thinking outside of the box, they are good at cookie cutter deployments as per a "blessed" checklist but outside of that I've found them to be quite useless.

A 3+1 CPG configuration meets the requirement and should allow for two disks to failures dependent upon density and sparing. More than that you would lose data because I think the minimum disks supported on the 8000 series is 6.

I think that markinnz put it best when he said "Well for the moment ignoring the god awful configuration you have been sold". This needs to be addressed and you should consider making HPE swap out those drives and/or SAN at no cost to you. If you chose to stay AFA I would recommend the following config:

Disclaimer I am not a VAR but an end user of the product.

3PAR 8450 (2 node)
x16 -3.84TB SSD's
This config will allow you to add 8 drives for expansion

OR

3PAR 8450 (2 node)
x1 -expansion enclosures
x32 -1.92TB SSD's (16 per disk enclosure)
This config wll still allow you to add 16 drives for expansion

If compression and remote copy is what is causing your high CPU you may want to look into the 4 node 8450. Another thing to consider, if the other 8200 is JUST an RC target and not serving production data you could just leave that one in place with its current config and demand that HPE provide you with a 4 node 8450. You have a lot of options and I would bet that HPE is pretty motivated to avoid bad press so use that to your advantage. Also, do not be afraid to throw your sales guy under the bus, it's easy to sympathize and relate to them as people but my God too often they propose some shitty ass solutions and we customers should not have to live with it.

_________________
Adam Berus - I.S. Tech Lead - Wolverine Worldwide Inc.
x1-7200, x2-7400, x1-7450, x1-8400, x2-8440
3PAR Technologies: AFC, AO, FilePersona, RemoteCopy (RCIP)