Which Dell RAID controller uses all drives for linear reading?

Hello,

I could not understand exactly what you mean by linear reading. Could you explain a little bit, please? I also share this page https://dell.to/2PE9H7J with you where you can check the PERC controller cards.

Let us know if it helps.

I/O operations can be:

linear read

random read

linear write

random write

 

My question about single linear reading:

On Dell PERC H730P, RAID10, Stripe 256KB, with 4*SSD SATA linear read show only ~800MB/s.

On a controller that reads from all disks speed should be ~2200 MB/s (4*550).

>I would refer you to the published spec sheets and manuals of the controllers

spec sheets and manuals do not have this info.

 

>I would not expect those mirrored drives to provide for a performance increase

Mirror not increase performance only on write operations. On read operations should be increased performance.

but I take it the Dell developers didn't realize it.

 

Hello,

 

I would refer you to the published spec sheets and manuals of the controllers. You should be able to locate them by running a search for the controller name and the term "spec sheet."

 

Because this is a RAID 10 and accounts for some mirroring, I would not expect those mirrored drives to provide for a performance increase, though.

 

H730P Spec Sheet: https://dell.to/3kvFW7b

 

PERC 9 User Guide: https://dell.to/33MwA14

Also found this issue on Dell PERC H730P, RAID5, Stripe 64KB, with 7*SSD SATA:

linear read show only ~940MB/s.

On a controller that reads from all disks speed should be ~3300 MB/s (6*550).

 

Are all Dell PERC models affected by this issue?

>Do you have a chance to try another benchmark

 

>Is this on the bare OS or a guest virtual machine?

Windows Server 2019 VM on ESXi 6.7, VMware Paravirtual controller

 

>Does the RAID initialization complete?

All tasks complete

Hello,

I don't familiar AIDA64 disk benchmark, but Do you have a chance to try another benchmark that is supported SSD(some programs could be supported only spinning disks) It may be due to test program used. You can also check the following to see if something else could cause. Is this on the bare OS or a guest virtual machine? Does the RAID initialization complete, if it doesn't or is still doing anything, such as rebuilding, initializing, if so please wait for let that complete and re-test. 

Hello,

 

We apologize for the delay in getting back to you on this matter. I spoke with a few colleagues from both the server platform and storage teams. This is an unusual question as the you seem to expect all disks to read at the same time. RAID doesn't work that way, as the data is spread across all drives for redundancy. The exception would be RAID 0, which does not offer redundancy.

I looked further into the linear read / raid issue and the only detail I could find was regarding SW raid using RHEL. Even these do not allow all drives to read at the same time. Here is a short passage on linear raid:

https://dell.to/31qhblr
Linear RAID
Linear RAID is a simple group of drives to create a larger virtual drive. In Linear RAID, the chunks are sequentially separated from one member drive and go to the next drive only when the first one is fully populated. This grouping does not provide performance benefits, as any I / O operation is unlikely to be split between member drivers. Linear RAID also offers no redundancy and actually reduces reliability - if any member drives fail, the entire array is unusable. Capacity is the sum of all member disks.

>>RAID doesn't work that way, as the data is spread across all drives for redundancy.

I've already taken redundancy calculations into account.

RAID10 of N disks have N*R read speed

RAID5 of N disks have (N-1)*R read speed

(where R is read speed of one disk)

 

let me show you this in the figure, which shows the distribution of data across the disks of the array:

as you can see, if you need to read one megabyte of data (S1+S2+S3+S4) from an R10 array, you can read:

S1 from disk1

S2 from disk3

S3 from disk2

S4 from disk4

since all stripes are on different disks, you can read simultaneously from all disks (ang get N*R read speed)!

a similar situation for an array R5 - you can read simultaneously from (N-1) disks (ang get (N-1)*R read speed).

 

 

>>Linear RAID

its simple JBOD+SPAN, and it has nothing to do with this topic

https://en.wikipedia.org/wiki/Non-RAID_drive_architectures

no answer?

Dell RAID controllers can't read from multiple disks simultaneously?

 

---

@maa1 wrote:

no answer?

Dell RAID controllers can't read from multiple disks simultaneously?

---

Sure they can, like all RAID controllers made in the last two decades can do (probably longer).

But the whole question makes no sense whatsoever, and I suspect there is a lot of misconception on your part. For some reason, you are concerned about linear reading (which is an operation where data which is located close by so it can be read without repositioning the head). I assume this concern is because of what you see (and how you interpret what you're seeing) from your benchmark tool. 

First of all, these benchmark tools work generally fine for desktop systems with single hard drives (or even a simple software RAID0 or RAID1), however they don't for hardware based RAID controllers with large caches. Simply because they can't see if read data comes out of the controller cache or from the disks, or if data that is written is only reported as written but still in the controller cache. Also, many controllers have different settings for things like caching, which all are completely transparent to the OS and thereby to the benchmark tool. 

Secondary, your assumptions completely ignore that, in a RAID config, the performance numbers of a single disk simply don't multiply. On top of that, you also ignore that spinning disks' performance varies notably by the area the head is over. 

Lastly, the benchmark screenshot shows that you are benchmarking a virtual disk, not a real one, and aside from the fact that the data location of the virtual disk has no relation whatsoever with where the data is present on the physical drives, there is also latency and other effects introduced by the hypervisor which means you can't reliably determine the performance of a physical storage solution in a virtualized environment. 

The problem with storage benchmarks is that it's a very complex topic with dozens of factors playing into it, and it's not enough to run some benchmark and look at the numbers.