VMware Cloud Community
XZim
Contributor
Contributor
Jump to solution

Nimble Storage

I'm thinking of investing in Nimble Storage to replace our current FC SAN infrastructure. It is does some pretty clever stuff with SSD and snapshots for backup ...however they are a relatively new player in the storage game with a small (but growing)client base.

Anyone using Nimble already and what has your experience been particularly with running VSphere on this SAN?

Reply
0 Kudos
68 Replies
XZim
Contributor
Contributor
Jump to solution

The 90 days mentioned on their website is a suggestion it is not a limitation. (I've re-confirmed this with Nimble).

The only limit, as with all storage, is capacity.

Reply
0 Kudos
Josh26
Virtuoso
Virtuoso
Jump to solution

JayL12 wrote:

For those of you using Nimble storage, because it only keeps snapshots for up to 90 days, what is your solution to keep your backup data for longer than 90 days?

Surely you're not planning on keeping your business data, and your only backup of that business data, on the same storage device?

Edit: Even without the 90 day limit, as per subsequent posts.

Message was edited by: Josh26

Reply
0 Kudos
JayL12
Contributor
Contributor
Jump to solution

I'm planning to get one at a DR site so they replicate.

Reply
0 Kudos
walter_09
Contributor
Contributor
Jump to solution

Does anyone have any updated input or experiences? I am currently evaluating options from IBM, EMC and Nimble. I have had a WebEx demonstration with Nimble and was thoroughly impressed. Like others have said, it sounds too good to be true.

And please only respond if you have more than 5 posts. I can tell which people are trying to market for Nimble. Its not hard to tell Smiley Wink

And by the way if you couldnt tell I am new to the forum, I signed up after finding this discussion on a google search Smiley Happy . So I also just wanted to say hi!:smileylaugh:

Thanks everyone in advance for your input!

Mike

Reply
0 Kudos
AllRegs
Contributor
Contributor
Jump to solution

I don't have any posts- but I found this on google as well. We've had a CS220 in production for about a year. It sounded too good to be true for us as well- but since we didn't have a lot of space to put a huge SAN we went with nimble and are very happy. There is one drawback for us though- and it is how Nimble handles thin provisioning, over-provisioning and volume cleanup. Generally- it doesn't clean volumes. Let me explain with a simple example:

Let's say you have a Nimble with 10TB of storage on it. You decide to add a large 6TB volume because you are doing a migration and you want to consolidate all your VMs using all the space temporarily and then you plan on copying most of the VMs out of the 6TB. Nimble will let you do this- but once data has been written in to the 6TB the Nimble never checks to see if the data is no longer there- there is no cleanup routine. This means that after you are done with this migration you must remove everything out of the 6TB and then delete the volume on the Nimble- that is the only way to reclaim space. Ideally you would put things on your Nimble on a more perminent basis- but this issue has gotten me in trouble a couple times in the last year.

With that said we saw another demo for a product that I feel would be an even better fit for my company - it's called Tintri (www.tintri.com) - I do not work for them- but their approach is even easier than Nimble (and Nimble is pretty easy). Their device is simply a large datastore. No volumes, volume collections, LUNs, Partitions... you get the idea. They have a device. The device is a large SAN datastore. Attach it to your vCenter and you are done.

Check them both out,

-Mike

Reply
0 Kudos
pizaro
Enthusiast
Enthusiast
Jump to solution

Hi Mike, first of all, thank you for being a Nimble customer!  I work for Nimble Storage...

The space reclaimation piece is now enhanced in 1.4 release.  You could get early access now through the support portal.  In this release, we have implemented VAAI Space Reclaimation primitive (a.k.a UNMAP).  With UNMAP, VM deletion or svMotion will trigger an UNMAP operation on the array side, cleaning up the unused blocks on the VMFS volume.  This should address the scenario you described.

HTH!

-Wen Yu

twitter: @wensteryu

blog: www.supersonicdog.com

Reply
0 Kudos
Yps
Enthusiast
Enthusiast
Jump to solution

Does some other guys got some more feedback on Nimble.

Have someone compared them with new vendors, like Tintri, PureStorage ?

How long time does a failover to second controller takes?

How much is performance degraded during firmware upgrade ?  I´m using EQL today, and performance are going down to < 100IOPS for like a minut during fw upgrading.

Is it easy to do fw upgrade? From gui or cli ?

Is the SRM integration good?

How are the performance during a rebuild of a disk? I know our old EQL PS400E went totally down during rebuild process for a day.

How granualar are perminssions in the array ? Could I setup different levels of admins ?

Have someone added a expensionshelf in production without taken the system down?

Reply
0 Kudos
sakman9999
Contributor
Contributor
Jump to solution

  • I have not compared Nimble to other newer storage vendors.  We purchased about one year ago and at that point in time there wasn't much credible direct competition.  If I were to buy today then I would look very closely (including getting demo units) at the two that you mentioned as well as comparable offerings from the big boys.
  • I haven't actually timed controller failover, but it's quick enough that our VMWare servers (and most Windows servers, except for a W2003 machine that just needs to go away) don't notice it.
  • I have not seen performance degrade during a firmware upgrade.  It updates the standby controller first, reboots that controller, and then makes it the active controller while to original active controller is upgraded.  Thus, the live active controller is never under additional load.  The only problem that I have had is with that W2003 server that just doesn't like to deal with the failover/cutover part.
  • Firmware upgrades are insanely easy to perform.  I have performed the last two from home.  I did get a little burned on the last one because I was aggressive in rolling it out and missed Nimble's attempt to block my installing of it.  Nimble discovered shortly after releasing the software that a specific issue affecting my unit (and those of others) would cause problems for the upgrade.  As a result, the upgrade was a little bit of a pain, but even so I completed the entire process from home and not once wished that I had stayed in the office to do it.  From what I have been told, both customers and Nimble folks learned a lot from that incident.  We did not suffer any downtime or performance degredation through the failed update attempts. 
  • I don't know about how SRM works with Nimble.
  • We haven't had to rebuild a disk, but given the architecture and metrics from our usage I don't imagine that there would be much of a performance hit.   Over the past 30 days our cache hit ratio has been in the 85% range, meaning that 85% of read requests were not serviced by the spinning disks.  Writes are initially cached to flash memory (before going to the spinning disks) and so I wouldn't expect to see a write performance hit during typial write operations.  Of course under heavy read or write operations during a a rebuild I can see how the caching wouldn't be enough to mitigate the rebuild activity.
  • Permissions on the array are not granular.  Nimble acknowledges this as a weakness and I believe that they are working to introduce a more extensive security model.  Ask your rep for details.  For us, it's not an issue but I can certainly see where it would be a problem in larger shops.
  • We have not added a shelf to it.

I strongly suggest getting a demo unit into your site for a spin, especially if you have historically dealt with more "traditional" storage players.  It's quite different from other approaches to storage, and I see how that differentiation can either be a good thing or a bad thing depending upon your expectations.  For us, it has been a great thing.

Hope this helps.

Reply
0 Kudos
stormlight
Enthusiast
Enthusiast
Jump to solution

Edit

If you find this or any post helpful please award points
Reply
0 Kudos
iamxCPx
Enthusiast
Enthusiast
Jump to solution

What's you guys take on Nimble isn't using an "Enterprise Class" SSD and instead just a standard "Desktop" SSD?

Will that affecting some performance issues?

On other thread they were saying this was a concern compare to Tegile array that's using enterprise class SSD.

Thanks for the feedback.

Smiley Happy

Reply
0 Kudos
lobo519
Contributor
Contributor
Jump to solution

While I think the Intel SSD drvies they use are the best "Desktop" SSD you can buy , Nimble's all around use of consumer grade hardware with the price they are asking was one of the larger reasons we choose not to buy

That was a year ago. We decided on an Equallogic 6100X and are quite happy with it.

Reply
0 Kudos
Yps
Enthusiast
Enthusiast
Jump to solution

Nimble don´t save any data on SSD, only use SSD as random read cache, with raid0.

All data are located on SATA with Raid6.

You can´t compare EQL with Nimble, the design is complete different.

The biggest concern I got agains EQL is the large blocksize, 15Mb, so snapshots/replicas will be hugh! Nimble got variable per LUN and compression.

I got around 100% changerate per day on our VMFS LUNs on our EQLs.

Reply
0 Kudos
lobo519
Contributor
Contributor
Jump to solution

I can and did :smileymischief:

I was simply  responding to the question regarding the hardware Nimble uses. I belive Equallogic to be superior.

Reply
0 Kudos
iamxCPx
Enthusiast
Enthusiast
Jump to solution

@Yps

Based on what their dashboard showed me, we don't have to configure RAID at all with Nimble. Which I kind of like that I don't have to worry about RAID 5, 6, 10.

So I'm confused. How do you know if they are using Raid 6?

Do you mind sharing? 

Reply
0 Kudos
lobo519
Contributor
Contributor
Jump to solution

Its Definatly RAID 6 - I just asked them

Reply
0 Kudos
Josh26
Virtuoso
Virtuoso
Jump to solution

CP wrote:

@Yps

Based on what their dashboard showed me, we don't have to configure RAID at all with Nimble.

Whether you configure it or not, it's going to be there. It simply defies logic to have multiple disks of any kind, with no RAID, whether it's in software, hardware or otherwise.

If any SAN hardware doesn't give you the option to choose your RAID level, it will be a very hard to take it to a tender where that sort of thing is part of an evaluation criteria.

Reply
0 Kudos
pizaro
Enthusiast
Enthusiast
Jump to solution

Disclaimer: I work for Nimble Storage.

Nimble does use RAID-6 - our architecture does write coalescing in NVRAM, then write sequentially to disk.  Therefore, RAID performance overhead is greatly reduced, without sacrificing the ability to protect against double disk failure.  There is no need to pick from a menu of RAID choices, and have to decide tradeoffs between performance, capacity, protection.  One thing you have to select with Nimble is the performance policy - we support variable block sizes for another layer of optimization.  For ESX with VMDK, simply choose the VMware policy of 4KB block size.  Remember this performance rule of thumb - you don't want the block size from the storage logical volume to be larger than the application block size.  When in doubt, go small.  If you don't know your application block size, you could go geeky and use vscsistat in ESXi (just google and you'll find good instructions).

RE: the use of MCL SSD from Intel..we don't use SSD as a tier for write - it's used for caching reads.  Doing so eliminates the RAID overhead as a copy of the block is either in disk/NVRAM.  Also, we eliminate write amplication effect by populating blocks sequentially (as you might know - you can't just free up a subset of blocks within a given page). 

RE: usage of commodity hardware.  It's not just about the type of hardware that is used...what matters the most is how the software layer monitors, detects and responds to hardware failure.  Don't overlook the importance of proactive health monitoring and auto support. 

Wen Yu

twitter: @wensteryu

blog: www.supersonicdog.com

Reply
0 Kudos
slyfoxxx
Contributor
Contributor
Jump to solution

Does anyone have a list of compatible switches to use in conjuction with the Nimble? I currently have an HP 8200 zl and or an Hp 3800. Do we need a higher grade switch?

Reply
0 Kudos
Josh26
Virtuoso
Virtuoso
Jump to solution

slyfoxxx wrote:

Does anyone have a list of compatible switches to use in conjuction with the Nimble? I currently have an HP 8200 zl and or an Hp 3800. Do we need a higher grade switch?

Whilst I don't know of a specific Nimble answer I can say that none of the storage vendors I've worked with qualify specific switches. Even HP, when they want you to buy their switch, only provide switch requirements.

Here is their requirements. It's exactly the same as Dell's, and good general advice, so likely quite appropriate for Nimble:

  • Network port must be enabled (apparently).
  • Gigabit – 10GbE is recommended in large environments
  • Fully Subscribed non-Blocking Backplane
  • 512KB buffer of cache per port
  • Flow Control
  • Jumbo Frames
  • RSTP or MST

Watch out with these, early HP switches advised they supported Flow Control and Jumbo Frames - but could not do both at the same time.

Reply
0 Kudos
eegolf
Contributor
Contributor
Jump to solution

We are a netapp partner and recently picked up Nimble. We switched out our Netapp 2040 with a nimble cs460. Nothing bad to say about either. I wrote a technical write up on our web site that goes into the technical information on our environment. http://www.ciosolutions.com/Nimble+Storage+vs+Netapp+-+CASL+WAFL

Reply
0 Kudos