Zeusram vs. Intel P3700

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
So I finally got all my stuff together to run this test.
Running on a FreeNas VM with 8GB RAM, 12 slow spinners behind the slogs (mirror'ed Toshibas), Ubuntu VM on NFS.
I was running the same benchmarks as Benjamin here (SSD ZFS ZIL SLOG Benchmarks - Intel DC S3700, Intel DC S3500, Seagate 600 Pro, Crucial MX100 Comparison | b3n.org).

I compared S3700 (200GB), P3700 (400GB), Zeusram (8GB)... and the winner is : Intel P3700

20 runs of each test, average and max values in the Charts

upload_2017-2-6_3-4-53.png

upload_2017-2-6_3-5-20.png

upload_2017-2-6_3-5-33.png

I was slightly surprised to see the Zeusram beaten by so much...
I have attached it with a dual port cable to a M1215 adapter so it should not be limited by that.
 

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
Have not run it this time but iirc something like 5 MB/s seqwr.
 

whitey

Moderator
Jun 30, 2014
2,770
865
113
38
Not surprised here, I'm impressed with how close the ZeusRam was hangin' in pretty tough/close on the first two tests. I can't quite follow your graphs x-axis...is that nearly 500MB/s on the P3700 acting as a SLOG on that spinner pool? What tool did you use to perform benchmarks? NM I see it was sysbench. Be interested to see if fio runs match up closely...they 'should'.

EDIT: Of other interesting note, if the run was only 30 seconds, I'd be Leary to call that a 'well burnt in' benchmark but good high level/baseline results to start for sure. Both the Zeus and P3700 have such insane endurance and low latency that I would expect them to keep up to snuff and have very little degradation.

DOUBLE NINJA EDIT: Now who wants to sell me a P3700 400GB model for $300 :-D hah
 
Last edited:

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
Have not run it this time but iirc something like 5 MB/s seqwr.
Actually they are around 15 MB/s seqwr, give or take. Might include full details in another post containing some more devices I have here.

I can't quite follow your graphs x-axis...is that nearly 500MB/s on the P3700 acting as a SLOG on that spinner pool? What tool did you use to perform benchmarks? NM I see it was sysbench. Be interested to see if fio runs match up closely...they 'should'.
Yes on sequential write it was MB/s and P3700 peak value was 501,08.

EDIT: Of other interesting note, if the run was only 30 seconds, I'd be Leary to call that a 'well burnt in' benchmark but good high level/baseline results to start for sure. Both the Zeus and P3700 have such insane endurance and low latency that I would expect them to keep up to snuff and have very little degradation.

DOUBLE NINJA EDIT: Now who wants to sell me a P3700 400GB model for $300 :-D hah
Sysbench was used with a 6GB file only (timeout 5mins), so no this does not measure consistent all day load.
I just wanted to get a feel for the performance, and in my home environment I most likely will not see all day load anyway;)

P3700 used prices are plummeting as well, saw several 400GB ones going for <€300 in recent weeks
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,004
1,569
113
CA
> 35M it appears to hit steady state: Intel SSD DC P3700 2.5" NVMe SSD Review | StorageReview.com - Storage Reviews

It's not so much the home work load works the drive out it's that once it's in steady-state the load is just enough to keep it there. This is where you should be testing ZeusRAM to others because the Zeus should not be slowing down where-as the P3700 eventually settles to around 170k from around 420k. The Zeus is rated at 100k and theoretically should perform there all the time, and at higher IOPs at lower QD than P3700 but we've yet to see any benchmarks put this to the test.

If you're using VMs and the Zeus or Intel are SLOG on your storage pull all the Guest OS on the VM are constantly doing some minor reading / writing so the SLOG will be hit while not hard, it should be continuous/regular access.
 

whitey

Moderator
Jun 30, 2014
2,770
865
113
38
Still waiting for your testing/results T :p

If I could just find a damn P3700 400gb for a reasonable price I would jump but alas none to be found w/ all the sharks arnd here smelling blood. :-D
 

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
Happy to run additional benches if you let me know what parameters you are looking for ?:)
Same size (6GB) just more iterations ? 200? 2000? ;)
Or larger size?
Or different benchmark? Which one? Which parameters?
I used the one from the link as I wanted to be able to compare the old and my results, but to specifically test Zeusram I am happy to run others as well.
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,004
1,569
113
CA
I would do whatever you need to push`em into steady state and see what the #s drop too for random writes / SLOG usage.
 

whitey

Moderator
Jun 30, 2014
2,770
865
113
38
fio/iometer w/ lengthy runs/burn-in time, at least an hour if not 4 or so. :-D

Thanks for the testing so far!
 

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
Will need to set that up on the box first.
Running 2000 iterations of the 6GB seqwr test on the p3700 atm;)

Edit: Stopped after 500 runs, no dropoff visible.
Runnin fio now in warmup (4k, 1hr)
 
Last edited:

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
It was the default from where I took the command:p
If you are happy with the metrics (write speed and iops only) I can get fio going on a write run:)
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,004
1,569
113
CA
You said it was 4K read only not sure how that simulates a SLOG though? Also, would be nice to know QD for the tests.
 

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
Yes you are right, noticed that when you mentioned it;) - as I said copied the command without thinking about it.
Reran the tests with horrible values today, then noticed one of my spinners had dropped out of the array with SATA errors
Not sure whether this impacted results or not so will have to rerun/verify.
 

Rand__

Well-Known Member
Mar 6, 2014
4,494
878
113
Can't rerun atm as the replacement disk also started having SATA errors and also the second replacement disk I had now has errors :eek:
So am keeping usage of the array as light as possible till i get those replaced.
And I seem I have to rethink my storage condition for my replacement disks :p Or test them more often;)