VMware Horizon Community
ebelange
Enthusiast
Enthusiast

Nimble CS220-X4 vs Netapp 2220 vs Equallogic PS6100S vs Tegile HA2100

Given the 4 above options for a 500 seat Vmware View environment with Windows 7 as the preferred guest OS, which option do you choose and why?

All solution are all the same price range with Equallogic being roughly $2500 more than everybody else but it offers an flash array. I also looked at Tegile and it was also impressive.

0 Kudos
15 Replies
ElevenB2003
Enthusiast
Enthusiast

You should throw Tegile's Zebi HA2100 into the mix as well.

http://www.tegile.com/products/zebi-storage-arrays

iSCSI, CIFS, NFS, FC in a single 3U box with deduplication and compression on the fly.

0 Kudos
ebelange
Enthusiast
Enthusiast

Thanks, will definitely check them out.

0 Kudos
eegolf
Contributor
Contributor

You can see this article we wrote for more information. http://www.ciosolutions.com/Nimble+Storage+vs+Netapp+-+CASL+WAFL.

The main question is what type of disk is behind the SSD on the netapp 2240. Are you putting SATA or SAS disks on the netapp? But before we talk about that lets first talk SSD. Nimble doesn't pay a raid penalty on their SSD's where as Netapp's flashpool will be in a raid set and hence pay a penalty. However netapp's deduplication for a VDI workload will minimize the actual block footprint used in the SSD read cache(flashpool) so the raid penalty is probably not a factor. Basically from a read perspective it is a matter of which vendor's read caching algorithm you trust more to maximize read cache hit rates. Nimble seems to be pretty focused on implying their algorithms do a better job at determining which data to prefetch into the ssd read cache but I'm not sure how you actually measure that against netapp to determine who is right and who is wrong for VDI workloads. We have a customer running about 600 VDI machines on a netapp 3040 with flashcache and deduplication seeing 90% cache hits continually which implies that something is working right with netapp. We haven't gotten a nimble deployed for a comparable VDI workload yet to make a comparison.

In terms of writes this is where nimble is much better than netapp...unless you pack the netapp with enough SAS disks to keep up. We have seen first hand how efficient nimble's casl algorithm is at converting random writes to sequential writes ...which is how it can be backed by SATA disk and still get such great write performance. The netapp write performance is going to be entirely dictated by spinning disks...with two exceptions.

     1.) The NVRam can soak up bursty writes...and drain those writes to the underlying disks.

     2.) WAFL is pretty good a making random writes perform faster than the actual disks spindles will allow if their is sufficient "Holes". Sufficient holes means the disk doesn't have to spin much to get to the next empty location to write the block. The likely hood of sufficient holes reduces overtime which is why a good netapp rep will size the system to handle the write IOP using normal disk IOP specs even through the system is likely to out perform that due to advantages from WAFL.

This point often is not fully understood but the workload the kills a nimble is one where the random reads "working set" required in a given period of time is larger than what fits in SSD and drops the cache hit too low. That puts that SATA disks in a situation where they have to switch context from read to writes too frequently and SATA disk pays a latency penalty for doing this context switch. You combat this scenario with adding larger SSD. The reality is that it is very hard to have a workload that does this(we haven't seen it yet) and nimble does give you very easy visibility into your cache hits. The fact that you are looking at an X2 already means you have nothing to worry about for your workload and honestly the X2 may even be more than you need.

If it were me, since VDI has problems if write IOP isn't good enough...I would pick nimble. But Netapp isn't a bad choice if it is sized correctly and meets your budget needs. Netapp also tends to have more software features...which they license...but can provide huge benifits when needed. Netapp also has the best implementation of NFS which can simplify administration.

I haven't used tegile but they seem to be playing the we are the cheapest card which maybe isn't a confidence invoking message for some. They use ZFS which was open source from Sun before Oracle shut down the party....ZFS - Wikipedia, the free encyclopedia..So you would have to have confidence in tegile's ability to maintain the code they didn't originally write. My only advice would be to make sure you find some customer references that make decisions similar to how you would with an environment similar to yours.  If you believe in the open source story and want to pay as little as possible then tegile seems to work.

Hope this helps.

0 Kudos
VBValtcho
Contributor
Contributor

WE have nimble s240 and I am running 800 link clones.220 concurrent users. No problems so far. We have 4 total nimble with SRM.

0 Kudos
Josh26
Virtuoso
Virtuoso

You are comparing a well established player in the enterprise, against a newcomer deploying consumer grade hardware.

It's not a fair comparison, but I think that demonstrates my answer.

0 Kudos
ebelange
Enthusiast
Enthusiast

You say you have the CS240, is that the base model or X2/X4. Nimble is quoting us the CS220-X4 which has 1.2TB of cache, I would hope it would be sufficient to run 500 Windows 7 desktop with 3-4GB of memory.

0 Kudos
ebelange
Enthusiast
Enthusiast

Netapp is putting 20 SAS spindles in the back end. My only concern is whether or not the flash pool will keep up with our workload. I'm being told by one of our resellers that these are 2 good solutions and that we will be happy with which ever solution we choose. The cost for both solutions is nearly the same coming both under $55K. So I'm really torn as to which solution I'm going to go with. Do I stick with Netapp which we've used for many years or do we go with a new comer which I hear all kinds of good things about... :smileyconfused:

0 Kudos
ebelange
Enthusiast
Enthusiast

Should I even consider Equallogic's PS6110XS?? That was the solution offered by Dell. From what i have read though, a lot of Equallogic customers have made the jump to Nimble so that's why we dropped them as an option.

0 Kudos
gmtx
Hot Shot
Hot Shot

I was an EQL customer and moved to Nimble for performance and value. Unless you're getting the deal of the century from Dell (and you make just get that right now as they appear to be in fire sale mode) I just don't think you'll be able to justify the cost/performance difference between the two once you start comparing.

Geoff

0 Kudos
eegolf
Contributor
Contributor

"Netapp is putting 20 SAS spindles in the back end. My only concern is whether or not the flash pool will keep up with our workload." Flashpool only matters for the read part of the workload. Any solution that uses SSD as read cache will perform well provided their is sufficient SSD space to handle the working dataset. In your case either solution will do fine for the read portion of the workload. The write portion of the workload will be spindle bound on the netapp ~ 20 SAS * 150 IOPS per disk = ~3000 IOPs sustainable but able to handle higher bursts due to NVRAM. The nimble's write performance is CPU bound 2 serries can do about 15,000 sustainable and the 4 serries can do about 30,000 sustainable. All of this is assuming random writes at 4K size which is the appropriate assumption for VDI. So if you assuming 10 IOPS per desktop which is considered a medium workload then you need to have 500 Desktops @ 10 IOPS per desktop =  5000 IOPS for you solution and you should assume a 20/80 read/write workload just to be safe. Assuming 90% read cache then  20% read *90% hit rate * 5000 IOPS = 900 IOPS soaked up by the read cache. This means your solution needs 4100 Write IOPS to handle the load. The netapp seems a little underspeced for the potential write load unless your VDI machines are using a ligher IOP per desktop load than we assumed. If i went netapp 2240 i would want about 36 SAS drives backing up the solution unless i knew for sure my IOP per desktop load would be small.

0 Kudos
VBValtcho
Contributor
Contributor

For our systems VMs we are running CS240G, which is perfect. We had the same for VDI but we were using too much CPU and cash. We did a live upgrade to CS440G with 2Tb SSD drives. It was an easu upgrade and the users did not notice at all. I took the CS240 to 28000 IOPs with no problem. In a normal day our system Array  hit around 5K IOPs and some time 20K. The VDI is running around 5-6K and I see spikes to 32K IOPS.

Here is my info if you need more. Cisco and Nimble have a design for VDI with UCS blades.

Valtcho.Bakalov@colorado.edu 3034923419


0 Kudos
VBValtcho
Contributor
Contributor

One more think we notice that 32 bit Windows 7 with 2Gb/2CPU is slower than 64 bit with the same CPU and Memory. Now we are migrating all VDI to 64 bit. It perform much better.

0 Kudos
DanSalazar
Contributor
Contributor

We currently run both an Equallogic PS6000 hybrid array (along with about 8 non-hybrid PS arrays) and a Tegile HA2100.  We're running about 200 desktops on each of the arrays along with several servers (Exchange, vCenter, View Manager, SQL servers).  We're seeing a very high percentage of cache hits on the Tegile and performance is noticeably better than with the Equallogic with very similar workloads.  This is backed up by what we're seeing in SANHQ and on the Tegile dashboard  After a lot of deliberation on which direction to go in the future we placed an additional order for another HA2100 last week. 

  • Deduplication and compression on the Tegile has allowed us to have a linked-clone storage footprint with persistent desktops with no apparent performance penalty.
  • Reliability and support with the Tegile has been excellent.
  • All of the essential features (snapshotting, replication, NAS functions) work very well and the user interface is easy to use.
  • Deduplication and compression functionality out of the box at no additional cost is a definite plus.
  • The ability to optimize lun performance based on the type of application is also very helpful. In our datacenter, the Tegile hosts a number of different types of applications and we can optimize the luns based on performance, compression, and deduplication.

The pricing for a new Equallogic PS6100XS was comparable to that of the Zebi.  Our only concern is the relative age of Tegile as a company.  Equallogic has served us well over the years but the advantages of Tegile at such a great price point outweigh this risk in our opinion.

0 Kudos
VBValtcho
Contributor
Contributor

Stay with Nimble and order 440 with 2TB cache. Make this storage only for VDI like us. We have Nimble for the servers and Nimble for VDI. Same for the redundant site as well. Nimble release 2.0 code and now you can bundle 2 storages together to share resources. I went to our vcenter, select 250 desktop and reset them. They reboot and the storage went to 160K IOPs. No sweat for the storage. Couple of minutes later all desktops were up like nothing happen. Their support is excellent. You want to have working VDI here is the recipe:

2 UCS chassis with 8 B230 M2 20 CPU cores and 512GB memory in.

Nimble CS440G-X4

EVGA zero client Tera2321 (latest one have 512MB memory) www.evga.com

VMware View 5.3 with remote experience enabled.

All this is 10GB network in the datacenter.

You will have 8 more slots for servers to grow, and with the current you can have 800 VDI easy if you allocate 4GB of memory per VDI. Add 8 more to it and you have 1600 VDI.

It is pricey in the beginning , but after that is just fun to sit and think where you will go fishing today Smiley Happy.

The installation will take 2-3 days at the most. I did the install many times.

0 Kudos
HendersonD
Hot Shot
Hot Shot

Another option to look at is Pure Storage. We purchased one of their arrays a little over a year ago and have been very happy. We run a mixed workload, 40 Windows servers (Exchange, SQL, etc) as well as 300 VDI desktops which will grow to about 420 by September. Besides IOPS which are important for VDI, latency is equally as important. The Pure with its all SSD design have consistent very low latency.

0 Kudos