Dell XtremIO Reviews

• Low latency
• High I/O
• Large bandwidth
• Helps my SAN perform at a high rate

It has improved the end-user experience:

• More fluid VDI
• Faster SQL searches
• Quicker data migrations

Replication: XtremIO has none

Controller failover: When you lose one of the scale-out nodes on XtremIO, your hosts will see a large drop in I/O while it occurs and EMC is unaware of how their controllers work. As it is shown in their demos, the work load after a failure is spread across each node. But if you look at the “ACTUAL” process, the work is loaded on each node until that node gets to 100%. Then, the work goes to the next node and repeats. So, in the event of a failure, you will have several nodes at 100% and others at 25% load.

Data recovery: In the event of a dual power supply failure, the array has to be recovered from a backup.

I have used EMC for two years.

We did have stability issues.

We have not had any scalability issues.

Technical support is average. EMC has no telemetry.

We switched for performance and cost.

The initial setup was straightforward.

There are better solutions available.

We evaluated Pure Storage, Nimble Storage, NetApp, IBM, and Cisco.

Don’t use this product. There are better solutions out there.

Data services: Deduplication and Compression. XtremIO performs both these data efficiencies in-line with no latency cost to the hosts. Also the APIs, using APIs you can provision an entire ESXi cluster using PowerShell and VC credentials within a few minutes.

I have leveraged XtremIO to reduce data center foot print significantly saving my company tens of thousands annually in floor space, power, and cooling.

Volume limits of 8192 per XMS and stability. I have unfortunately had some bad experience with bugs and failed upgrades. It has improved with the new 4.0 code but it’s still not as good as typical EMC.

XtremIO and Pure Storage have a volume limit. XtremIO is 8192 logical volumes per XMS management server. XMS can manage multiple XIO clutters, if you do that your volume count is 8192 across all the clusters XMS is managing. Pure Storage has just released code to go to 5000. A logical volume on XIO is any volume, whether it’s a snapshot or a volume presented to a host. Whereas Pure Storage only counts a snapshot that is hydrated.

Two years.

See my comments above regarding room for improvement.

Not yet.

I have had pretty good support including being escalated to engineering very quickly, this could be considered a good thing or a bad thing…

Mostly all VMAX, we switched for the consolidation of datacenter space.

It’s more complicated than a VNX but easier than a VMAX. The fact the array has a separate management server and that server has its own code levels that are not in line with the array can be confusing, much like Solutions Enabler. I highly recommend always purchasing a physical XMS and only use the virtual appliance in a pinch.

If you plan on replicating XIO to XIO you will have to buy RecoverPoint, XIO to XIO only needs SE RecoverPoint lic but that won’t allow you to use RPA to migrate from VNX/VMAX to XIO so consider the EX lic. Remember ever snapshot counts as a logical volume so every replication snapshot counts toward your overall 8192 volume limit. I highly recommend large LUNs to keep your volume count low. We are using 4TB or larger.

I POC’d XIO, Pure Storage, and IBM SVC with TMS array. We chose XIO because of its scalable by adding bricks where as Pure Storage you have to buy more arrays creating islands and more management. Also, had some issues with PURE post processing of data services. We felt the SVC for IBM introduced significant latency to their AFA TMS. TMS alone is the fastest All Flash Array but it doesn’t have any data services, that is what SVC is used for.

Consider your migration methods to this array and know your data. For instance, if you are running Windows 2012 you may consider disabling ODX on the entire array which is disruptable or disabling on every single 2012 host. Talk to your SE about this.

Snapshots, deduplication, compression – We are using the XtremIO for multi-version databases, in this respect being able to snapshot a database consistency group to create DEV/UAT versions while utilizing deduplication/thin provisioning/compression allows us to maintain the numerous copies of each database we need. To go along with that, being able to refresh any snapshot set with the contents of any other snapshot set (in the same lineage) allowed us to reduce our refresh times from hours to minutes.

DEV/UAT refresh periods reduced from hours or longer to under five minutes.

Volume count. A hard limit of 8192 volumes per cluster. This becomes an issue with DR replication and RecoverPoint and trying to maintain the best RPO possible.

There is a limit of 8192 volumes/LUNS that can be created. This includes all volumes/LUNs presented to hosts along with all snapshots, so it becomes very easy to bump up against the limits in certain circumstances. For us, we use RecoverPoint to replication between XtremIO devices, and since RecoverPoint creates a snapshot of each volume to allow for point in time recovery that results in a lot of snapshots that have to be accounted for.

One year.

None, the product has been working as expect without issue.

None that weren’t expected. This is a scale out product, not scale up or scale up and out. You can go from 1 XIO brick to 2, 2 to 4 and then 4 to 8. If you know this up front it is very easy to plan around.

Extremely good. EMC has been outstanding with support, especially when using their call-home utility ESRS.

Yes, we had a previous all flash array vendor, however we encountered many issues with support, scalability and a general lack of data efficiency services that ultimately were more important than all flash performance.

Initial setup is very straightforward. There is a configuration workbook you complete to provide the basic information (IP addresses, domain names, mail, SNMP, etc.) and work with a Dell EMC project manager to get it installed. Array comes preconfigured from a storage standpoint, so once it is up and running you can start allocating storage immediately.

Not much to a basic XIO installation, everything is licensed initially. There is no built in replication or other business continuity features, if you need that you will need to look at products like VPLEX to sit in front of the XtremIO.

Pure Storage and Violin Memory.

Understand your workloads and use-cases. This is not a perfect solution for all flash workloads. If you cannot take advantage of deduplication and compression there may be better/cheaper solutions. If you want simplified replication, this is not the product for you. For us, performance wasn’t the prime driver. We wanted a scalable solution and our workloads could take advantage of deduplication extremely well so this was an obvious choice.

• Integrated Copy Data Management (iCDM): It allows instant creation of high performance copies by accelerating the environmental testing with no need for redoing builds. A duplicate copy can be spawned up for DEV- trunk/ Integration/Test with QA.

• Virtual Desktop infrastructure (VDI): Where one consolidated infrastructure that leverages off resources can act as a baseline and be locked down with a secure image to all call center agents.
• Rapid Inline Deduplication/ Compression: Helps capacity planning to reduce the amount of SAN needed. Prevents the deduplication of the same data. This can allow ratios of 2:1, or even as high as 7:1, depending upon what data subsets are used.

• Utilizing the snapshot technology, (recovery point), has been great for development.
• Inline data reduction using deduplication and/or compression has improved batch times for one of our PostgreSQL database environments.

• The management server needs to be integrated. XtremIO VNX and VMAX have separate management software stacks for managing the various arrays.
• We would also like to see one universal management view into all these sub-systems. For example, IBM SVC and Pure Storage Purity each have one universal software management view.

I have used it for just over three years.

The earlier version before Version 4.0 had several stability issues. It is now more stable, albeit not 100%.

There were no scalability issues.

I would give technical support a rating of 6/10.

We have a two vendor approach with HDS and EMC.

The initial setup was straightforward, providing that the correct code levels were applied before initializing.

When considering pricing and licensing issues, note the following:

• The 8192 volume limit
• Snapshots count as logical volume
• Every replicated snapshot also counts

We evaluated Pure Storage, but on a FA-420 controller.

A testing and integration PoC needs to be done with full diligence.

Its consistent speed and replication features are most valuable to us. It provides a better user experience and replication for DR purposes.

In the VDI arena, it helped speed up VDI sessions and also helped improve the amount of VDIs versus disks.

Replication: EMC should have native replication ability, not something you have to take an outage for in order to install, i.e., if you did not purchase the replication when you rolled out the storage array.

Native replication is important for a few reasons:

• If something goes wrong, it gives you more of a "one throat to choke" scenario, meaning you only have to talk to one or two vendors.
• Native replication is usually cheaper, especially IP replication (versus fiber-based replication).
• When replication is native, you can usually have more insight at to what is replicated and what is not, instead of having to query various different things to get an answer.
• Native replication is usually easier to set up and maintain. It also has little less administrative work and overhead costs.

I have used this solution for about two years.

I would give the technical support a 8/10 rating.

We had VDI running on another storage array. Thus, we switched to this solution so that we could utilize the speed and the inline deduplication.

Make sure you understand and check how EMC’s replication and licensing policy for replication works.

Also make sure you are at the latest code and fully understand how the grouping and clustering of hosts on the storage array works.

• Speed of operation: We had several SQL databases that pounded EMC CX4-480 and EMC VMAX 20K.
• XtremeIO handles the flow well, running on the controller memory, rather than straight to the drives.

It has offloaded high IOPS processes and cleared the main arrays for bulk work.

Even with the fast SSD drives and processing on the controller, there was still a lag on the FC ports.

The initial node came with only two FC ports per controller. It was used for multiple ports on the VMAX to spread traffic over several VSANs.

For more detail:

I had 4 dH2i powerpath servers hitting it, along with 4 vmware clusters 8 host each, on a X1 brick we only had two controllers both with 2 port
So a total of 4 FC ports.

Compared to the VMAX 20K, where I had 8 ports on vlan 2, 6 ports on vlan 100, 8 ports on vlan 50, so I was able to spread the traffic around between process.
I had 2 directors on one VMAX, whereas I had 3 directors on the other VMAX.

With only 4 ports on the xtremeIO, the most I could do was send traffic on 2 ports to two different VLANS one on each controller.
So my comment was get additional ports, so the DH2I servers don’t hog all the IOPS.

Recommend getting the second brick X2 and the matrix switch, then with 8 FC connector can start spreading the traffic.

The company had me routing the data thru a fabric switch MDS9500, separate from the main traffic as this was a test.
Most of production was on 4 other MDS9500 switches.

Monitor of the switch, did not show a bottleneck going to the servers, only on the 4 8GB FC going to the XtremeIO.
Connect to different blades on the 9500.

Don’t think they have touched it since I left. Nor on the other 8 SAN units.

We have been using the solution for two years.

We had some stability issues. Initially, one of the ports failed. The unit could not use a LUN larger than 2TB. After testing all our variables, it was determined that it was XtremeIO, and a patch was created.

The servers were attached with both PowerPath and VMware 5.1 datastores, via a MDS 9500 Fabric Switch network.

It didn't expand to the second Node X-2, although that was a stated option.

The technical support was poor, even during the port or 2TB limit. It was rare to hear back from the technical analyst looking at the unit from ESRS.

Over my thirty years in the IT field, I have tried many solutions. I worked with:

• NetApp
• EMC SANs
• Direct attached SCSI drive units
• An IBM 4300 unit attached by VMware 2.5

Compared to others, the setup and operation is easy. I worked at the company almost three years, learning XtremeIO with little assistance from co-workers or the vendor.

Even before Dell bought it, EMC pricing was steep.

We evaluated Pure Storage and NetApp.

Our company didn’t send anyone to operations training until we had the unit for two years. I would advise you to send your technical expert to take the training early on.

This product was the best fit for our 5000+ active VDI environment. It also works well with database use cases.

The speed is extremely valuable. It is blazing fast. The inline deduplication is amazing and the snapshot capability is very useful.

• The physical architecture could use some higher levels of redundancy.
• The past upgrades were highly impactful to active workloads.
• Previous levels were also susceptible to security vulnerability scanning.

There were stability issues with upgrades and general protection against data center power outages.

I did not encounter any scalability issues.

I never had any issues with support on the storage array.

We never ran VDI on any other storage platform, although we did have any other options available: NetApp, 7-Mode/cDOT, VMAX.

The racking and initial setup was performed by EMC services. This is the normal case for this product. The first two setups were not installed properly. However, the other five installations were perfect.

I was not active in this portion of the product deployment.

I never saw any active workloads on other products.

I would make sure you do not put all your eggs in one basket on this hardware.

I would make sure updates have an extended down window and are prepared for the array to not service I/O during these updates as a worst case scenario.

It deduplicates well, but I would not run a database and the log files on the same array, to be safe.

• Dedupe, because we achieve 3.5X space savings with XtremIO
• Snapshot copies
• Ease of use
• Improvement in storage efficiency without compromising on performance

We have just used the product in a proof of concept test and we are yet to move it into production.

Integration with other EMC SAN storage array ecosystems like VMAX3, Unity, etc., especially for consistency group protection: This will help to consistently recover applications which are spread across multiple product lines like VMAX3 and XtremIO for various reasons.

I could give an example of SQL DB hosted in EMC VMAX array with SRDF replication to a remote site for disaster recovery.

If we have to move part of the volumes (like TempDB, Indexes,logs. Etc) into XtremIO and part remains in VMAX, the whole server won’t be able to able to get recovered with consistent point in time of recovery due to different replication technologies in 2 different platforms (VMAX and XtremIO).

I meant to say that it may be perfect for other environments but not for environments which are heavily dependent upon SRDF based replication. There might be ways to get around like implementing Recoverpoint for VMAX and XtremIO but the integration was complex in our use case.

I have been using the product for one year.

We have not encountered any stability issues.

We have not encountered any scalability issues.

The technical support is good.

We did not use any other solution beforehand.

The initial setup was straightforward with a very simple-to-use interface.

All the software-inclusive licensing simplifies the cost model.

We did not evaluate other options.

Mileages vary, so please test the dedupe rate and performance metrics you are achieving for your workload, before implementing in production.