I recently switched fom OMV to Rockstor, and I was wondering about a small performance issue. I’m using the exact same hardware for the NAS (Atom D510 with 2 GB RAM and a 3gb WD Red drive on a Windows SMB share). When I copy into and out of the NAS from my desktop HTPC (an SSD drive) I would get about 80-90 MB/s with OMV, and I get about 55 MB/s with Rockstor. I realize they use different file systems, but I was wondering if there is a setting in Rockstor that I need to tweak. Since I’m not really a power user, I’ve just used each distribution out of the box (no changes made in configuration to either from the defaults). Any ideas on things to check?
Do you have compression turned on for the rockstor array? Could you also list all of your hardware it’s possible that OMV loaded a different driver for your NIC.
Good morning (my local time) Gregory @gcole and welcome to Rockstor Community!
Reading your post made me remember my experience (and panic) while comparing simple samba share on a linux box (140-150 MB/s on Gigabit net) vs freebsd box samba share (30-40 MB/s with strange jumps to 90-100 and back to 10-5 MB/s too)
On my current Rockstor box (same net, gigabit cards) average speed is the same of the linux box (both with read and write to the share), so if you’re always on the same hardware it’s a little bit strange compared to OMV.
As suggested by @Spectre694 that should be nic driver related, specially with a Realtek nic cards.
I had a similar problem on the 140-150 MB/s linux box (with a Realtek nic card) after a kernel upgrade, driver got updated to r8169 (official kernel driver) but i had to downgrade to r8168 (realtek official driver)…and that box is still running with that old r8168, because with r8169 fails (kernel panic).
Realtek nic cards are strange animals
From a shell on you Rockstor box try lsmod to list your kernel loded modules and let us know
Thanks for the reply Mirko. In order to avoid issues with Realtek drivers, I used an Intel gigabit card (PWLA8391GTLBLK PRO/1000 GT) instead of the onboard LAN. Here is what an lsmod generated for me:
My current lsmod (on dev env and not on a running stable/testing channel box) differs from your specially for iptable references (maybe dev env differs from running box, but don’t think so)
lsmod shows e1000 which is the driver for that card.do a
dmesg | grep ‘Ethernet’
just to check for sure. Also is the HTPC a windows box? If so go to the wired driver settings and turn off a setting called Large Send Offload (LSO) and test again.
For whatever reason no one including Intel has made a NIC card where that wasn’t broken and Windows enables it by default. On mine disabling that changed my speeds from around 50MB/s to 100MB/s ish.
Here are the iperf results. The first is with my normal setup, which is the PC and the Rockstor running thought a Trendnet Gigabit router. The second test was with both the PC and the Rockstor connected straight into the Tp-Link gigabit router. Not any real difference to speak of. Any chance it is testing the write speed of the USB boot device rathe than the WD Red storage drive? Just a thought as to why iperf was even lower than my transfer speeds into the Rockstor share.
@gcole I have just popped a “PWLA8391GTLBLK PRO/1000 GT” (same model as you detail) in a Rockstor 3.8-13.03 and got the following:
Rockstor machine running iperf -s (ie server mode) it’s ip is 192.168.1.247
Client is Fedora 23 using a build in RTL8111 running iperf -c 192.168.1.247 -i1
Client connecting to 192.168.1.247, TCP port 5001
TCP window size: 85.0 KByte (default)
So there’s something fishy going on re the sporadic drop to 672Mbps but the rest looks about right and nothing like the very low figures you experienced.
And in the opposite direction as I have often found this test to be asynchronous:-
ie running now iperf -s on the Fedora 23 and iperf -c client-ip-address -i1
So given no jumbo frames and around 94% efficiency the max should be around 118 MB/s so a little short but fairly usual measurements here with that card in a Rockstor machine, I also have a few things going on with the client that could be dropping these figures a tad but they are in the ball park of what is expected I think.
Regular gigabit network with both machines attached to an HP ProCurve.
Just turned off some network stuff on my Fedora 23 client and ran this time through a silent network TP-Link Archer C5 v1.2’s build in switch.
This section with desktop running iperf -c to the -s on Rockstor with the intel GT in and we have:-
So now consistent and a better result than before. The reverse is pretty much the same as before.
And just for comparison I ran the same tests but this time with the lesser Realtek RTL8169C in the Rockstor machine:-
Again on a silent network through the TP-Link’s build in switch.
iperf -s on Rockstor machine first
as can be seen the performance is significantly less with this very common realtek chip, at least it used to be pretty common.
so to summarise:-
Intel PRO/1000 GT card in Rockstor - 851/877 Mbps
Realtek RTL8169C in Rockstor -636/582 Mbps
Note that the Intel GT card is a desktop class network card not a server class.
I have posted these results as I’ve been meaning to test this routers switch again and given your post and my finding the same model network card here it seemed like an opportunity to provide context.
The Rockstor machine was a single core Celeron with 2MB ram.
Thanks for the testing Philip. I am using the exact same router myself (TP-Link Archer C5). I dug up an old system here in my office and I’ll put Fedora on it and see what I get. Also grabbed a few new cables, just to check the physical layer as well. I appreciate the data points.
Interestingly enough, the issue seems to be with the Intel cards under Windows. I threw together another Rockstor on an old Lenovo desktop with a Broadcom gigabit onboard NIC. Here is what IPERF returns between the two Rockstors:
[root@rockstor ~]# iperf -c 192.168.1.105
Client connecting to 192.168.1.105, TCP port 5001
TCP window size: 85.0 KByte (default)
[ 3] local 192.168.1.115 port 43934 connected with 192.168.1.105 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-10.0 sec 1.01 GBytes 868 Mbits/sec
So I guess the next step is to start adjusting things under Windows and see what happens there. To take the hardware out of the equation, I’ve also used another Windows box that was a steady 38 MB/sec. I also switched out the router and switch and cables, so I’m fairly sure the answer isn’t within those parameters.
Seems to clearly be the Intel NICs under Windows causing the problems. I took out the Intel NIC and used the onboard Realtek LAN and got this into the Rockstor:
That’s more like it! So now I have to figure out what the issue is within the Intel NICs that is causing them to slow down. I would think Intel to Intel would be ideal, but something doesn’t add up. The problem is my HTPC has onboard Intel. The Windows machine I just used to get the correct transfer rates is a backup server running Windows 10 and didn’t have Intel as the onboard LAN. As suggested earlier, I tried disabling LSO but didn’t see any change. Any other suggesting to check? Again, my Rockstor and my HTPC both have Intel LAN chips
