Linux File-System Benchmarks On The Intel Optane 900P SSD

[leipero]

why don't you inform your misconceptions?

My only misconception is what others (and you) already mentioned, everything written above is true, that was my undesrstanding. It is possible that metadata does not get updated when data is updated/refreshed/checked, however you could say that in your post I relied now so I could know, it's much faster and easier than reading pages about how data is manipulated by ext4 (and others) file system. If i ever need to know that, I will read it ofc., till then, simple reply with corrections is enough. If you think there's something else I said that might be misconception feel free to correct me.

[nomadewolf]

nobody needs zfs in kernel anyway, its design is obsolete

I happen to agree, but if it's actually impossible, maybe it's easier to accept...

[nomadewolf]

Reading is fundamental.

Correct.
And maybe you should do that also...

I mean, if you already know that there are licensing issues... isn't it pretty much all said?...

[nomadewolf]

Oracle recently talked about changing ZFS' license so it could get in the Linux kernel.

Let's hope they do.
It'll at least remove a lot of unnecessary discussion.

[Niarbeht]

Correct.
And maybe you should do that also...

I mean, if you already know that there are licensing issues... isn't it pretty much all said?...

I made that post because your post was literally quoting my post, and then effectively a sentence from my post.

Your entire post was a reply to, and repeat of, my post. Which is kind of strange. Why would you repeat something I wrote?

[nomadewolf]

I made that post because your post was literally quoting my post, and then effectively a sentence from my post.

Your entire post was a reply to, and repeat of, my post. Which is kind of strange. Why would you repeat something I wrote?

I kinda just read your first sentence, and replied to that...

[Niarbeht]

I kinda just read your first sentence, and replied to that...

*glares*

[vladimir.mijatovic]

I know it's not the newest post, but since it's about Intel optane 900p, it fits here better then other forums.

I ran some FIO benchmarks with Phoronix test suite on my server with Proxmox 5 and Ubuntu 16.04 LXC container.
Proxmox and container are running on Intel Optane 900p, 480GB in PCIe slot.

LXC showed signifficantly slower performance (208k IOPS vs. 94k).

HOST:
OPERATING SYSTEM: Debian GNU/Linux 9
Kernel: 4.13.13-1-pve (x86_64)
Compiler: GCC 6.3.0 20170516
OS on EXT4 filesystem.

Flexible IO Tester 3.1:
pts/fio-1.11.2 [Type: Random Read - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 4KB - Disk Target: Default Test Directory]
Type: Random Read - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 4KB - Disk Target: Default Test Directory:
Average: 813 MB/s
Deviation: 2.65%
Average: 208333 IOPS
Deviation: 2.64%


CONTAINER (Ubuntu 16.04 LTS):
Flexible IO Tester 3.1:
pts/fio-1.11.2 [Type: Random Read - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 4KB - Disk Target: Default Test Directory]
Type: Random Read - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 4KB - Disk Target: Default Test Directory:
Average: 369 MB/s
Deviation: 2.15%
Average: 94200 IOPS
Deviation: 1.75%


Also, benchmark results on bare metal host are app. 30% worse than those that Michael got with his tests.
Is anybody else noticing similar results?
Any advice what might be the cause?

Thank you.

[polarathene]

HOST:
OPERATING SYSTEM: Debian GNU/Linux 9
Kernel: 4.13.13-1-pve (x86_64)
Compiler: GCC 6.3.0 20170516
OS on EXT4 filesystem.


Also, benchmark results on bare metal host are app. 30% worse than those that Michael got with his tests.
Is anybody else noticing similar results?
Any advice what might be the cause?

Thank you.

I've just been going over the article and discussion. Just to be clear, you're benchmarking the Optane drive with the running OS installed to it? That can affect your results. You'll notice on the first page Michael touches on the test system specs and has another drive in use, not only the Optane.

Furthermore, some results may be impacted by the other hardware(it's running a beefy AMD EPYC with 32 cores 64 threads at 2.2GHz and 128GB of RAM.

You'll also notice that on the software side of things, he's running Ubuntu 17.10 with 4.14 kernel and GCC 7.2, you're running Debian/Proxmox with older kernel and GCC, the newer kernel could introduce improvements to things like filesystems and maybe the newer GCC has some optimizations that affect any compiled software? Debian/Proxmox might have some distro specific differences to Ubuntu as well.

Proxmox would be running some extra services and might have different config to suit running the host as a hypervisor more?(I dunno).

Can't really comment on the LXC container performance, Phoronix didn't run any of the benchmark via a container. Are you not seeing overhead(disk or others) using LXC with other applications?

You could look into your disk I/O scheduler(probably should be noop for the optane?), your CPU frequency governor(setting to performance might help if it's not).