1. Which one is the fastest (architectually)?
- Pretty sure the 3PAR would be here. First we have dedicated asics for moving data onto disk. This allows us to do non-impact full thin provisioning, inline zero detection and inline zero reclaim. Have your HP sales folks give you a Ninja Stars output which is generated from real world testing to give you an idea of performance. Plus we just announced inline deduplication for the 7200 and an All Flash Array kit for the 7200 for SSD's which is starting a 35K (street) so that controller has some oomph to run as an AFA. Plus we are a true activeactive architecture from the ground up, not alua based so you don't have to worry about controller balancing etc. As far as I know we pretty much smoke everyone else especially when you start adding things like thin-provisioning etc. Plus 3PAR is VMWmware's reference architecture so we are ready for vvol's when they launch. Basically the key is the ASIC. It offloads so much work that it leaves the main CPU for doing the interesting software features. Now I will be honest all the vendors will give you numbers and tell you we are fastest and they are slowest. And you can pretty much find a workload where your array will excel and theirs will fall apart, or worse case (I have seen EMC do this several times) give you their benchmarks of a competitors system. The problem is they will configure it wrong and make a competitors system look horrid.
Being as unbiased as possible though from what I have seen the 3PAR is the fastest mostly due to wide striping, VNX can't keep up because it's not really a wide stripped array and when you start enabling features and then drops off quickly (they are trying to get VNX to match 3PAR's) and finally Compellent is just weird. They do this weird write logging thing, which is neat in practice but the challenge with tricks that I typically expect them to work to a point in which they fall apart so you have to ask yourself what happens to the system when you hit the higher points of the architecture, how does it react to changing workloads.
2. Which one has the best auto-tiering?
- So 3PAR has been doing this for ages. It is called Adapative Optimization. It is basically hands off automated tiering. It will analyze over time and move blocks up and down based on useage. You can do three tiers. The moves are based on "regions" of 128MB. Care should be taken in respect to IO density. Assuming you get system reporter you can actually run reports before you add tiers to help you size properly based on your workload. If you have to buy upfront then best practice is to do about a 60/40 Tier 1/Tier 2 split because we are very cautious about NL drives impacting performance. However we just announced adaptive flash cache which extends the cache to SSD's. It's early but I beleive that by allowing a much larger SSD cache, will help alleviate some of the potential issues you can have in an improperly sized tiered architecture. Note though that the newly launched 480GB cMLC drives are allowed to be used Adaptive Flash Cache, just MLC drives and the 1.9TB cMLC drive. Also there has been a lot of work with Storage Federation, we are really early on in launching features under "Peer Motion" but that will essentially be adding transparent migration of workloads between arrays. Say different tiers of arrays like an All Flash 7450 and midrange 7400...
3. Is one next-higher model of a series absolutely worth buying? (Here with consideration of costs please )
- So there are two big enhancements you get when you go from the 7200 to the 7400, however those are only realized when you from the dual to the quad controller setup.
1) You can go to 480 drives instead of 240 drive
2) You get persistent cache - what that means is when you have a controller failure you do not go into write through mode. This is absolutely essential in mission critical environments as a controller failure will not impact performance. This is part of the basis for the 6 nines guarantee we offer.
3) You do get more than double the performance but you only really see this with a lot of SSD's
4) The 7400 license cap is higher so that impacts cost - a lot!
I would say get both priced and see what the delta is. If you think you need to go 480 drives someday then start with a 4n 7400. 4nodes does make upgrade a little more expensive because best practice is to purchase drives for both controller pairs. So you upgrade in larger chunks. Honestly I would not say it's a "required" upgrade just a consideration based on your environment. Heck our converged system 700 uses the 7200 not the 7400 so it's a useful box and we probably sell more than any other model
1. Which one is the fastest (architectually)?
- Pretty sure the 3PAR would be here. First we have dedicated asics for moving data onto disk. This allows us to do non-impact full thin provisioning, inline zero detection and inline zero reclaim. Have your HP sales folks give you a Ninja Stars output which is generated from real world testing to give you an idea of performance. Plus we just announced inline deduplication for the 7200 and an All Flash Array kit for the 7200 for SSD's which is starting a 35K (street) so that controller has some oomph to run as an AFA. Plus we are a true activeactive architecture from the ground up, not alua based so you don't have to worry about controller balancing etc. As far as I know we pretty much smoke everyone else especially when you start adding things like thin-provisioning etc. Plus 3PAR is VMWmware's reference architecture so we are ready for vvol's when they launch. Basically the key is the ASIC. It offloads so much work that it leaves the main CPU for doing the interesting software features. Now I will be honest all the vendors will give you numbers and tell you we are fastest and they are slowest. And you can pretty much find a workload where your array will excel and theirs will fall apart, or worse case (I have seen EMC do this several times) give you their benchmarks of a competitors system. The problem is they will configure it wrong and make a competitors system look horrid.
Being as unbiased as possible though from what I have seen the 3PAR is the fastest mostly due to wide striping, VNX can't keep up because it's not really a wide stripped array and when you start enabling features and then drops off quickly (they are trying to get VNX to match 3PAR's) and finally Compellent is just weird. They do this weird write logging thing, which is neat in practice but the challenge with tricks that I typically expect them to work to a point in which they fall apart so you have to ask yourself what happens to the system when you hit the higher points of the architecture, how does it react to changing workloads.
2. Which one has the best auto-tiering?
- So 3PAR has been doing this for ages. It is called Adapative Optimization. It is basically hands off automated tiering. It will analyze over time and move blocks up and down based on useage. You can do three tiers. The moves are based on "regions" of 128MB. Care should be taken in respect to IO density. Assuming you get system reporter you can actually run reports before you add tiers to help you size properly based on your workload. If you have to buy upfront then best practice is to do about a 60/40 Tier 1/Tier 2 split because we are very cautious about NL drives impacting performance. However we just announced adaptive flash cache which extends the cache to SSD's. It's early but I beleive that by allowing a much larger SSD cache, will help alleviate some of the potential issues you can have in an improperly sized tiered architecture. Note though that the newly launched 480GB cMLC drives are allowed to be used Adaptive Flash Cache, just MLC drives and the 1.9TB cMLC drive. Also there has been a lot of work with Storage Federation, we are really early on in launching features under "Peer Motion" but that will essentially be adding transparent migration of workloads between arrays. Say different tiers of arrays like an All Flash 7450 and midrange 7400...
3. Is one next-higher model of a series absolutely worth buying? (Here with consideration of costs please )
- So there are two big enhancements you get when you go from the 7200 to the 7400, however those are only realized when you from the dual to the quad controller setup.
1) You can go to 480 drives instead of 240 drive
2) You get persistent cache - what that means is when you have a controller failure you do not go into write through mode. This is absolutely essential in mission critical environments as a controller failure will not impact performance. This is part of the basis for the 6 nines guarantee we offer.
3) You do get more than double the performance but you only really see this with a lot of SSD's
4) The 7400 license cap is higher so that impacts cost - a lot!
I would say get both priced and see what the delta is. If you think you need to go 480 drives someday then start with a 4n 7400. 4nodes does make upgrade a little more expensive because best practice is to purchase drives for both controller pairs. So you upgrade in larger chunks. Honestly I would not say it's a "required" upgrade just a consideration based on your environment. Heck our converged system 700 uses the 7200 not the 7400 so it's a useful box and we probably sell more than any other model
การแปล กรุณารอสักครู่..
