Probably not going to matter for VMware workloads, but enterprise storage does have some very significant advantages over the mid-range storage. 1. Reliability, we are talking about 99.999 vs 99.9999. Try doing a firmware update for a storage that serves some mission critical app and you will quickly realise how scary failing over the controller can be. 2. QoS, typically mid range storage allows a single LUN to go as fast as possible, saturating the entire array. Enterprise storage don’t allow any of these, no single LUN can hog the entire array. 3. Symmetrical, typically you have active-passive and active-active for mid range boxes. On enterprise array, this is where you start seeing symmetrical active-active. These array has huge interconnects between the controllers and ensure that no matter which port you hit, it’s always an active and preferred path. Path thrashing that leads to random lockups doesn’t quite happen here. 4. Drive quality, ssd on the more expensive boxes tend to be a grade higher than the cheaper boxes, you see SLC and e-MLC on the more expensive boxes and cheaper TLC on the cheaper boxes. Then again, most people probably will never need any of these. I won’t recommend an enterprise box for VMware related workload. Thou I’m leaning towards AFA like Pure and XtremeIO.
Thanks for that comment. I totally agree with the observations but at the same time I have some contradicting viewpoints as well. Gives me an idea to discuss some of these points perhaps in my next video. Thanks again for that awesome insight.
In a way, VMware runs really well on both mid range and enterprise storage and most of the VAAI/VASA stuff are well supported, unlike mainframes which only run on enterprise storage. The expectations and demands of VMware workload are very different from the workload that demands enterprise storage such as a super critical database. VMware workload are typically “atomic” in nature consisting of many small VMs which often average out to maybe 50 iops or so. This allows easy scaling out instead of needing a scaling up. If you ever run out of storage space or iops, you can easily just add another array and start spreading out the load via storage vMotion. This is rather different from your super critical DB, if your controller run out of performance, you really want some way to scale up with minimal downtime. QoS is well taken care by SIOC, pathing is well taken care by SATP and it’s easier to spread out the vms across the controllers, meaning many of the enterprise storage feature are not really that important for VMware workloads. The expectations for uptime is also quite different here. There is more leeway for downtime and workload which are more critical can usually be safely vMotion off before any big maintenance. Feature set is also quite different, enterprise storage are more geared towards reliability whereas mid range storage nowadays are heavily geared towards efficiency. You get various tiering and data reduction capabilities which is more useful here.
@@MooEy84 Totally agree with your comment and hence my thoughts precisely. High end enterprise subsystems today are still relevant for certain usecases and mainframes is one such instance. On the point of a physical hosts completely saturating a storage controller, thats definitely a real issue but with most mid range offerings, there are often multiple tiered offerings as well. If you had a requirement that potentially sucks up utilisation on the controllers, some of the "higher" tiered offerings within the mid range space today will address it comfortably... Virtualised workloads are primarily the dominant workload in most datacentres, and whats left of physical workloads are candidates for re-platforming moving forward. This brings up an interesting discussion. Do you build a setup catering for the dominant tech, less dominant tech, or each to its own? I personally believe, in a balance for cost, performance and efficiency, a 1 storage to catch all isn't the best of strategies. Not for now at least...
I think you have got your concepts wrong fundamentally. The performance that you see on VMWare comes from the storage. VMWare has been built for utilization and not for performance. So the concept of LUN hogging the controller, as well as SSD comparisions are not correct as well.
Another outstanding video by Charles. In a future video, I'd be interested in seeing price per TB ranges between Expensive, Mid, and Cheap. Thank you to Charles for being generous with his time and knowledge.
![[Gegagedigedagedago] NEW Help Gegagedigedagedago Nugget escape from Nikocado Avocado Challenge](http://i.ytimg.com/vi/hJWnwRC2Fz8/mqdefault.jpg)
Probably not going to matter for VMware workloads, but enterprise storage does have some very significant advantages over the mid-range storage.
1. Reliability, we are talking about 99.999 vs 99.9999. Try doing a firmware update for a storage that serves some mission critical app and you will quickly realise how scary failing over the controller can be.
2. QoS, typically mid range storage allows a single LUN to go as fast as possible, saturating the entire array. Enterprise storage don’t allow any of these, no single LUN can hog the entire array.
3. Symmetrical, typically you have active-passive and active-active for mid range boxes. On enterprise array, this is where you start seeing symmetrical active-active. These array has huge interconnects between the controllers and ensure that no matter which port you hit, it’s always an active and preferred path. Path thrashing that leads to random lockups doesn’t quite happen here.
4. Drive quality, ssd on the more expensive boxes tend to be a grade higher than the cheaper boxes, you see SLC and e-MLC on the more expensive boxes and cheaper TLC on the cheaper boxes.
Then again, most people probably will never need any of these. I won’t recommend an enterprise box for VMware related workload. Thou I’m leaning towards AFA like Pure and XtremeIO.
Thanks for that comment. I totally agree with the observations but at the same time I have some contradicting viewpoints as well. Gives me an idea to discuss some of these points perhaps in my next video.
Thanks again for that awesome insight.
Just out of curiosity, why would it not matter for VMware workloads?
In a way, VMware runs really well on both mid range and enterprise storage and most of the VAAI/VASA stuff are well supported, unlike mainframes which only run on enterprise storage.
The expectations and demands of VMware workload are very different from the workload that demands enterprise storage such as a super critical database. VMware workload are typically “atomic” in nature consisting of many small VMs which often average out to maybe 50 iops or so.
This allows easy scaling out instead of needing a scaling up. If you ever run out of storage space or iops, you can easily just add another array and start spreading out the load via storage vMotion. This is rather different from your super critical DB, if your controller run out of performance, you really want some way to scale up with minimal downtime.
QoS is well taken care by SIOC, pathing is well taken care by SATP and it’s easier to spread out the vms across the controllers, meaning many of the enterprise storage feature are not really that important for VMware workloads.
The expectations for uptime is also quite different here. There is more leeway for downtime and workload which are more critical can usually be safely vMotion off before any big maintenance.
Feature set is also quite different, enterprise storage are more geared towards reliability whereas mid range storage nowadays are heavily geared towards efficiency. You get various tiering and data reduction capabilities which is more useful here.
@@MooEy84 Totally agree with your comment and hence my thoughts precisely. High end enterprise subsystems today are still relevant for certain usecases and mainframes is one such instance. On the point of a physical hosts completely saturating a storage controller, thats definitely a real issue but with most mid range offerings, there are often multiple tiered offerings as well. If you had a requirement that potentially sucks up utilisation on the controllers, some of the "higher" tiered offerings within the mid range space today will address it comfortably...
Virtualised workloads are primarily the dominant workload in most datacentres, and whats left of physical workloads are candidates for re-platforming moving forward. This brings up an interesting discussion. Do you build a setup catering for the dominant tech, less dominant tech, or each to its own? I personally believe, in a balance for cost, performance and efficiency, a 1 storage to catch all isn't the best of strategies. Not for now at least...
I think you have got your concepts wrong fundamentally. The performance that you see on VMWare comes from the storage. VMWare has been built for utilization and not for performance. So the concept of LUN hogging the controller, as well as SSD comparisions are not correct as well.
Another outstanding video by Charles. In a future video, I'd be interested in seeing price per TB ranges between Expensive, Mid, and Cheap. Thank you to Charles for being generous with his time and knowledge.
Thanks David! Appreciate the kind comments. I would love to get a chance to play with Wasabi at some point. Ive heard great things.