SOLVED CIFS over WiFi is very slow only to FreeNAS - Sanity Check

[jgreco]

At $10 per port I find it hard to rationalize cheaping out by going lower than the GS108T v2, but that's just me...

 

[yokken]

I have no clue what you are doing wrong, but I just did this:

View attachment 12112

[root@freenas] ~# cat /etc/version
FreeNAS-9.10-STABLE-201605240427 (64fcd8e)
[root@freenas] ~# sysctl net.inet.tcp.cc.algorithm
net.inet.tcp.cc.algorithm: cubic
[root@freenas] ~# sysctl net.inet.tcp.mssdflt
net.inet.tcp.mssdflt: 1460
[root@freenas] ~# sysctl net.inet.tcp.recvspace
net.inet.tcp.recvspace: 613200
[root@freenas] ~# sysctl net.inet.tcp.sendspace
net.inet.tcp.sendspace: 613200

They work for me. Unfortunately this system is not local, so I have no way to test wifi performance. You must reboot to apply the changes though.

So yeah, *definitely* can and does load the cubic kernel object and the settings *do* get set properly.

I wanted to give you a serious shout-out and huge thanks for this. Using the tunables shown in your screenshot, followed by a reboot, I am now able to push ~50MB/s sequential writes to my array over 802.11ac with the same ~60MB/s reads I was getting before the tunables. Amazing.

Backstory: I have a home lab with FreeNAS as an ESXi VM (passed-through M1015 w/ 3x 6TB WD Purple in RAIDZ1 and 8GB reserved ECC) and everything that needs fast access to the two main arrays is a VM on the same ESX host, so I'm not worried about the newreno optimizations over the cubic algorithm. Nearly everything else major that I do with the arrays is over my 802.11ac AP (Ubiquiti UAP-AC-PRO, excellent AP) with my Windows 10 workstation and my late-2013 rMBP. The rMBP has very solid 802.11ac and was able to hit ~60MB/s reads and writes with my old Netgear R7000, but when I moved to a Mikrotik CRS125 + Mikrotik NetMetal 5 (3x3 setup) I was unable to get those same speeds, often dropping to 30MB/s reads and 12MB/s writes. I was upset with the NetMetal because it wasn't cheap, at $150, but I also didn't buy a proper antenna, instead using the 3 that came with my R7000. Either way, it seemed low, and I did everything I could on the AP side to fix it, since speeds seemed great until I changed my router/AP setup, so I assumed it was a network issue. Fast forward almost a year, and I'm fed up. I decided to bite the bullet and buy a Ubiquiti AP after reading nothing but great reviews, opting for the UAP-AC-PRO, which was about $50 more than the Lite, but I wanted the top of the line so I got it. After setting up the UniFi server in a CentOS tool server VM I was running, I set up two separate SSIDs/networks, with one limited to 2.4GHz and the other limited to 5GHz. The 5GHz AC network on the Ubiquiti got me up to ~60MB/s reads from my FreeNAS array to my Windows 10 workstation as well as my rMBP, which was awesome for pulling down an ISO or game. I still couldn't push files to the array any faster than 15MB/s, though. Your tunables fixed this for me and now I am enjoying glorious AC goodness.

TL;DR: Couldn't get reads faster than 60MB/s and writes faster than 15MB/s with 2 different APs. Applied your tunables, now getting 60MB/s reads AND writes. Thank you so so much!!

 

[cyberjock]

@yokken

You are welcome. I'm glad they worked for you. ;)

 

[rungekutta]

Just wanted to add that I also set cc to cubic as suggested further above and it made a very significant difference. Now getting >300Mbit transfers over CIFS over WiFi (802.11n) which is a vast improvement. A little googling also revealed that Linux uses this algorithm per default from kernel 2.6.19 and above. Perhaps this would be something to consider for FreeNAS also...? Or at least a standard option in the interface somewhere (i.e. optimise for wired only or for mixed environments).

 

[hervon]

Applied Cyberjock tunables with success too. Before would get 40 MB/s max on AC. After reached 75 MB/s with consistent 60MB/s. Thanks!

 

[Stux]

Found an interesting page on FreeBSD (and other OS) tcp tuning while looking into this subject:
https://fasterdata.es.net/host-tuning/freebsd/

 

[Chris Moore]

Honestly, I just read through this thread and all I see is "yep.. wifi is still slow.. what's new?"

I don't really see any compelling data that anything is wrong. That is besides the fact that you may feel that Wifi should be faster. But that's also why somewhere we tell people not to complain about poor wifi performance because there's no fixing that fro this forum. :p

You certainly won't be the first to be unhappy with the performance of Wifi, and I have no doubt you won't be the last. Not even the last for this week.

This is the kind of response that makes people not only move on to something else, but also talk bad about FreeNAS. The guy has already shown that the wireless speed is faster to a Win 10 host and to a Linux host. It is only slow to FreeNAS. That's a problem with FreeNAS, obviously. You can not always point and say that it is something else.

This should be a bug report either to FreeNAS or to the TrueOS development team.

Sent from my SAMSUNG-SGH-I537 using Tapatalk

 

[anodos]

This is the kind of response that makes people not only move on to something else, but also talk bad about FreeNAS. The guy has already shown that the wireless speed is faster to a Win 10 host and to a Linux host. It is only slow to FreeNAS. That's a problem with FreeNAS, obviously. You can not always point and say that it is something else.

This should be a bug report either to FreeNAS or to the True development team.

Sent from my SAMSUNG-SGH-I537 using Tapatalk

You're right. If adjusting the cc algorithm improves performance for FreeNAS users without any downsides, then it should be a bug report / feature request. I mostly use FreeNAS is a professional context with a wired-only network. Wireless clients are on a separate isolated network segment, and so I can't really comment on the performance gains of cc_cubic.

However... if you've seen improvement from the tuning advice in this thread, please make a bug report. My experience is that the developers are open to suggestions. They'll hear you out even if they decide not to make changes. :)

Painting in broad strokes, I think it is probably better to by default tune for good performance with 1GB and wifi while documenting through tooltips (or something) steps that help tune for 10 gigabit and high performance. The people trying to extract maximum speed from their NAS will be more likely to be able to adjust what's under the hood than noobs.

 

[Stux]

You're right. If adjusting the cc algorithm improves performance for FreeNAS users without any downsides, then it should be a bug report / feature request. I mostly use FreeNAS is a professional context with a wired-only network. Wireless clients are on a separate isolated network segment, and so I can't really comment on the performance gains of cc_cubic.

However... if you've seen improvement from the tuning advice in this thread, please make a bug report. My experience is that the developers are open to suggestions. They'll hear you out even if they decide not to make changes. :)

Painting in broad strokes, I think it is probably better to by default tune for good performance with 1GB and wifi while documenting through tooltips (or something) steps that help tune for 10 gigabit and high performance. The people trying to extract maximum speed from their NAS will be more likely to be able to adjust what's under the hood than noobs.

I would agree and suspect that it's more important for the average user to get 100% of their peak wifi performance rather than 40% over wringing out a few gigabit extra from his non existent 10gbe LAN ;)

I plan on testing these tunables myself eventually. I'm curious if they provide good performance on both wifi and gigabit, with perhaps a regression on 10gbe.

 

[anodos]

I would agree and suspect that it's more important for the average user to get 100% of their peak wifi performance rather than 40% over wringing out a few gigabit extra from his non existent 10gbe LAN ;)

I plan on testing these tunables myself eventually. I'm curious if they provide good performance on both wifi and gigabit, with perhaps a regression on 10gbe.

If you do this, please do some benchmarks and get good data to put into a bug report.