RAID0 take that much CPU power?

Hi friends.
Did a little check today in order to estimate how RAID0 (2x840pro) affect the CPU (4960X).
I used 8 standard tests of CDM 3.0 with CPU thread activity window.
The results surprised me a little.


















Seq Read = 2%-CPU 2-threads active.
Seq Write = 2%-CPU 2-threads active.
512K Read = 2%-CPU 2-threads active.
512K Write = 2%-CPU 2-threads active.
4K Read = 2%-CPU 2-6?-threads active.
4K Write = 5%-CPU 2-6?-threads active.
4K-QD32 Read = 22%-CPU 3-threads almost fully active + minor activity in 5 others.
4K-QD32 Write = 21%-CPU 3-threads almost fully active + major activity in 4 others.


Any enlightenment?
How much those reads\writes affect the CPU without RAID array?

10X

 

What platform was this on? Is there a dedicated RAID card?

The CPU will push the components as hard as they can go and, vice versa. Especially on the higher queue depths.

Synthetic Reads (and no processing following...) from the storage subsystem is easy on the cpu; it is waiting for more.

Synthetic Writes on the other hand are the cpu pushing data to the drive(s) and the better the storage subsystem (and program used) the harder the cpu works.

 

Clevo P570WM / Sager NP9570. No hardware RAID.

Keep in mind that SSDs are too fast for high QD in real world usage in a consumer environment.

 

Too fast? Even though I know your point, I have to disagree.

Anything past level 1 cache is too slow...

Besides, the synthetic high QD on CDM is real world for some.

 

Running a database server, high QD, yes. Pr0n collection, high QD, no.

 

Thank you all for the educated answers.
Sorry for the delay.

Ill change my question a little.
Is there any way to isolate the RAID management process CPU Utilization in real time?

 

That's normal.

Have you ever tried loading a game while using something like MSI Afterburner to check out CPU usage? I bet you it'd be 100%.

The CPU has to work to do the workload. I've been running RAID0 (2 x 840 Pros) on my previous P55 (i7 870) and my current X99 (i7 5820k) no problem.

Just enjoy the speed.

 

It matters if you build large projects 10GB+ over and over. (Video games development)
The building process is multi threaded and uses 80-100% CPU Util.
I need to know because i don't want to take 4-6 threads (20%+ CPU Util) from the build process to the Raid Management algorithm.

 

In that case I'd say you'd be better off with a hardware RAID card.

 

My first reply in this thread has already answered your question. This isn't a guess. It is based on how computers and storage subsystems work.

If you are not using a dedicated RAID card, synthetic reads are giving you the best indication of the overhead the raid process requires re: cpu utilization. Writes mostly depend on the host cpu (whether using a dedicated RAID card or not).

In either case, make either (cpu or storage subsystem) more powerful and up to a point (depending on the real world workload presented), the cpu utilization will just keep going up.

For your current platform/components and for a synthetic workload (i.e. no work done afterwards with the data the cpu is reading/writing), the cpu utilization for the RAID is around 2-5% for lower queue depths.

For high queue depth (QD32) random r/w's, it gets to 22%.

In real world workloads which also significantly taxes the storage subsystem, I would expect the RAID array to be utilizing 10% to 15% average of your cpu. On the same workloads that are supremely pummeling the storage subsystem into submission; I would not be surprised to see 30% cpu utilization or more.

What I have found over the years is that anything over ~70% cpu utilization does not translate into significantly higher productivity on the programs I use most (this is a clear indication for me to consider a new platform). Which indirectly means that with a RAID setup capable of such high performance you are only using ~40% of the processor's capabilities for the task on hand (not a good trade off for most workflows - RAW 4K video processing excepted).

This doesn't mean that the last 30% of a cpu's performance are not important, on the contrary. To me, it means that a new platform running at ~70% or less on my existing software gives me an overall more productive platform than the old one running at 100% cpu utilization for a single program. For example; when converting RAW NEF files from my last shoot I don't want to dedicate a workstation to that task alone (although I can... I have many systems available that I can use...). I want to be able to continue sorting, renaming, and further editing already edited files - without slowing down the main task at all (converting the RAW NEF's) - in addition to checking my email, forums, creating invoices on my accounting program and updating my notes as needed.

To reiterate, getting an answer for "Is there any way to isolate the RAID management process CPU Utilization in real time?" depends on the actual workloads you run.

To really see what that % utilization of the RAID array is:
boot with your RAID setup and run your test (your specific real world usage is highly recommended here) and record your results. Now, reboot and select a non-raid O/S drive with the exact same O/S and software configuration and run your identical test again and compare your results. The difference is the RAID cpu utilization for that particular workload (and I would bet it will be in the ranges I indicated for most, real world workloads).

(Note:
While synthetic benchmarks give you a 'score' to compare with; they are no match for real world software that actually produce something at the end of the day.
Running your usual workloads is always the best bet of optimizing your system/process for your needs.
And if you don't need to multitask for certain stages of your workflow, getting the task done faster (period) is all that matters in the end - sometimes a RAID array will help with that, sometimes it won't. It all depends on the actual software you are running.

And why it is imperative to test specifically with it when deciding for highest performance/productivity options or processes to set).

 

Thank you very much (again) for your effort.
And for teaching me a thing or two.

I understand the importance of checking the real world workloads in both cases.
But i thought ill find a way to estimate it without disassembling my RAID array.
Ill wait with it until ill have to reinstall my system.

10X a lot.

 

Glad to help in a small way.

But no reason to reinstall the current setup - simply buy/use another SSD/HDD and ensure it mirrors the installation of the RAID array. Power off the system, remove the power from the RAID array drives and connect the power to the single drive and boot from it.

After you have your real world benchmark scores, simply shut down, remove the single drive from the system and connect the drives for the array. If, the results show that having a RAID setup in your use case is of benefit, of course.