SOLVED Slow LAN speeds (via FTP) since upgrade

Status
Not open for further replies.

cyberjock

Inactive Account
Joined
Mar 25, 2012
Messages
19,526
Ah, ok. That should be a valid test.

Well, based on your iperf tests, you have a network link somewhere that is 100Mb and not 1Gb. So its time to start hunting for it.
 

Whattteva

Wizard
Joined
Mar 5, 2013
Messages
1,824
It's certainly possible to do so through the Internet.
Whether or not is it wise to do so, is the more important question, since FTP traffic is all transported in clear text.
 

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
I wouldn't want to share via the internet (aside from the security concerns) mainly because my service is only 50Mb, which would be even slower than my current speeds. My main problem at this point is finding out where the weak link(s) is/are.. As far as I can tell, even using LAN Speed Test Lite, it's still reporting just under 100Mb/s speeds, which shouldn't be possible considering my setup. Interestingly enough, the connection between the two Win 7 machines (host and guest VM) registered just under 800Mb/s, so my problem obviously lies outward. My switch and router report Gb connections at the top end, so luckily my house wiring isn't the suspect. Perhaps Amazon duped me and sold me Cat5 instead of Cat6 somewhere.. Either way, I really appreciate your help Cyberjock - consider me educated.
 

cyberjock

Inactive Account
Joined
Mar 25, 2012
Messages
19,526
Few things to think about when trying to find the culprit:

1. "Green" switches and routers often don't give you the performance you'd expect. They save power by not having great performance. It seems stupid to sell Gb switches that don't actually do Gb speeds, but pretty much all "Green" switches, "Green" NICs, and Realteks shouldn't be expected to provide you the full bandwidth they claim. Don't like it, don't buy them. :P
2. Cat5 generally should work for Gigabit, unless you have very old Cat5 or cheap Cat5 cabling. Some stuff from China isn't real Cat5, but they'll gladly market it as such.
3. If your Cat5 runs near fluorescent lighting or other electrical wiring that has high noise(such as cabling attached to motors like refrigerators, vacuum cleaners, garbage disposals, etc.) that can cause problems.
4. Some Realteks just sucks. So try to rule out your desktops as being the performance barriers since they are very likely to have Realteks. If you question them at all, get an Intel NIC. I use only Intel in all of my machines despite all of them having onboard Realtek NIC(s). I learned that lesson back in 2009 and I've never regretted having Intel NICs.
5. Firewall software(Symantec in my case) really kills network performance for some connections but not others. So you might want to try disabling your Firewall software temporarily just to make sure its not the culprit.
6. The category rating of the cabling should be printed on the cable sheath so you can verify it is cat5(5e and 6 is recommended for Gigabit). I've seen people have problem with those female-to-female adapters as well as wall adapters.

Other than that.. good luck!
 

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
Odd that you should mention the "green" switches, as I definitely have one such TrendNet 5-port between the 2 Windows machines and the wall plug-in. As far as the Cat5 cable goes, that shouldn't be an issue in the first place, as I wired the whole house myself with bulk Cat6 from Monoprice (I also did it the right way, with a patch panel and punchdown Cat6 keystone jacks). I also took great care to keep my runs as far away from any flourescent lighting or other electrical (as the runs traverse the ceiling of my garage, going up into the ceiling to the floor above, this was something I had accounted for in my original planning).

The firewall suggestion is good, I'll try that next (and I'm a little ashamed I hadn't already tried that). I only use Windows firewall, FWIW, but I'm sure Avast has got its fingers in that pie, as well.

Lastly, are Realtek NICs really that bad? This forum is the first place I've seen totally rail on them, but if it's not just personal preference, I'd definitely like to hear more about that. Being a networking student and a freelance IT guy, I'm always interested in learning useful things. I also appreciate you excusing my inexperience with FreeBSD, as I have (up until I started using FreeNAS) always been primarily a DOS/Win guy, so I'm kinda learning as I go.
 

cyberjock

Inactive Account
Joined
Mar 25, 2012
Messages
19,526
Well, it's like this in the Realtek versus Intel(and Broadcom's are pretty good too apparently).

1. The CPU does all of the heavy lifting with Realteks. Intel's do most of the hard work on-chip. This doesn't matter if you have a 24 core system. But for those of us trying to make do with less than outrageously expensive hardware, you are going to have a hard time getting full 1Gb speeds from Realteks. Most people hit about 650Mb/sec max with Realtek. Go to Intel and instant 850Mb/sec+. Not to mention their driver is single threaded, so expect horrible performance with Intel Atoms and other "Green"(and old) CPUs. The more powerful the system the less improvement you'll see with Intel. But really, who cares? Intel's just work!
2. Intel has spent far more money to give great performance with low CPU utilization where Realtek is meant to be cheap so that they can make money off of the high quantity of network chips they sell.
3. Realtek doesn't really support their non-Windows driver, so they're written by programmers that have less than nice things to say about their hardware. If you want a good laugh, check out the programmers comments from prior versions here.
4. Plenty of people have had very bizaare performance and reliability issues with their network gear. No explanation can be found and they're beat their head into the walls until they give up. The buy an Intel NIC from ebay, and suddenly all of their weird issues are instantly gone.

Personally, I had performance issues between my server and desktop at home in 2008. I was a single user and very unhappy with the very poor network performance I was getting. 50MB/sec max for a single user when the hardware could do over 700MB/sec with its RAID array was disappointing. I tried tweaking the system registry and all sorts of stuff. Nothing really worked despite weeks of trying everything. Lots of people swore off Realtek and praised Intel, but I didn't see a point since 1Gb should mean 1Gb, right? Wrong. As soon as I replaced the Realteks in my server and desktop with Intels I got 100MB/sec. At that moment I swore off Realtek and I've never gone back.

In 2012 when I started getting involved with non-Windows stuff(mostly Linux Mint and FreeNAS) everyone said the same thing. If only I had known that a few Intel NICs would have mattered that much i would have started using them years before. But basically, when I go looking for new hardware, I always either go with Intel onboard or get an Intel NIC. I keep 2 spare Intel NICs around the house too just in case I need one for someone's computer. And for their price its kind of hard to say no considering how much performance improves by using them.

So yeah, basically those that are more seasoned will tell you Intel is just the shiznit. Less seasoned users might just dismiss potential limitations of their network as a bottleneck from somewhere else. But by far and large, the vast majority of "network problems" threads created in this forum are solved just by getting an Intel NIC. It's quite depressing because the forum is full of threads, most of which have the exact same solution. If people would just read and spend the $20 for an Intel NIC they could have saved themself a lot of time and the forum from "yet another non-Intel user that couldn't listen". :P
 

joeschmuck

Old Man
Moderator
Joined
May 28, 2011
Messages
10,970
This thread says "Solved" and I'm curious what solved it.

And as my friend states, RealTek performance is going to be based on your CPU horsepower. Intel does it's processing onboard. I have a mixture of both and in my systems it doesn't make a big difference which I use but I am using the Intel NIC in my FreNAS machine since I bough two of them for testing.
 

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
It wasn't really solved in a typical sense, just basically been narrowed down to a (probable) infrastructure issue. I assume a mod changed it to solved.

Speaking of which, the green TrendNET switch between is the only remaining source of the slow-down that I haven't been able to rule out yet. I'm going to cut it out of the line and re-test as soon as I get a chance, and I'll post the results here.
 

joeschmuck

Old Man
Moderator
Joined
May 28, 2011
Messages
10,970
Hum, not sure why a Mod would have changed the status, that is for you to change. Yes, please let me know what happens when you cut out the TrendNET. I have one but it's not in-line with any heavy traffic device. I do have a Rosewill Gigabit switch that I did test and it passed gigabit speed without any perceivable slowdown during my RealTek vs. Intel NIC testing.

So I just did a short set of tests using Iperf and a FreeBSD 9.1 VM, results were not what I'd expected. using the command on my FreeNAS (FreeBSD was the server using "iperf -s"):

"iperf -c 196.168.1.48 -r"

I got 386Mbit/sec. Swapping the server and client around didn't change the results by much. I suspect my VM is causing problems so maybe I'll check out a Windows version of iperf at a later date.

Using the examples you had, I got similarly very slow results as well.
 

cyberjock

Inactive Account
Joined
Mar 25, 2012
Messages
19,526
I don't think I've seen a trendnet "Green" switch tested here yet. But my Netgear "Green" switch only gets about 600Mb/sec. It's good enough for its location in a spare bedroom.
 

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
I'd be just fine with 600Mb/sec haha. The program I linked to a few replies back works well for Win machines (although there is a Win version of iperf floating around), I'd recommend that for testing from Win machines, just for ease of use. I'm still trying to get around to cutting that switch out of the mix. With the final episode of Breaking Bad having just started, homework, and working on clients' machines, I'm regrettably occupied at the moment.

BTW joe, my FreeNAS machine isn't a VM, just my seedbox Win 7 machine.
 

joeschmuck

Old Man
Moderator
Joined
May 28, 2011
Messages
10,970
BTW joe, my FreeNAS machine isn't a VM, just my seedbox Win 7 machine.
I completely understood, as was mine. I have Windoze 7 on a frankly killer machine running VMWare Workstation 8 and the actual VM was of a fresh FreeBSD 9.1 installation. I have to install iperf but that was about it. I plan to install a Windoze version of iperf so I can take the VM out of the issue because I know I have great transfer rates with my network but I was shocked iperf didn't show that so I seriously suspect the VM.

I will post my results once I get there.

EDIT: I wonder if one of the flavors of Live CDs has iperf on them.
 

paleoN

Wizard
Joined
Apr 22, 2012
Messages
1,403

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
I just ran the tests again with the green switch taken out of the mix. Still stuck just under 100Mb/sec. Any other ideas?

And joe, try this utility (it's what I'm using, and gave me the same test results that iperf did): http://www.totusoft.com/downloads.html

ETA: Just did some tests on another machine (which has a Realtek Gb PCI NIC) from the same switch my main Win 7 PC is hooked up to. Got 675Mb/s up and 300Mb/s down. Not great, but a hell of a lot better than 100Mb/s. So, this just got even more confusing.
 

joeschmuck

Old Man
Moderator
Joined
May 28, 2011
Messages
10,970
Brian,
Here is my posting for testing throughput.

http://forums.freenas.org/threads/intel-nic-vs-realtek-nic-performance-testing.10325/

But I did do a simple throughput test by copying a large file (5GB movie) from my NAS to the Windows 7 machine and got an average of 90MB/sec. Renaming the file to something completely different and copying that file back to the NAS got me a peak of 110MB/sec but average of 92MB/sec. Keep in mind that my throughput results dropped when I change from a RAIDZ1 to a RAIDZ2.

Do this kind of test to see if you are chasing your tail before going further with iperf. This will test the CIFS protocol unless you have NFS (must have Windoze Ultimate) working then I suspect it will be a little faster.

I need to look into iperf more and I'm hoping the Ubuntu Live CD has iperf. That would put my on a level playing field and I wouldn't be using a VM on my Windoze machine.
 

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
When I try to run the dd command from your post, I get: "dd: tmp.dat: Read-only file system".
I tried to mount it as RW, but it's telling me no such directory exists (I changed the directory to reflect actual folders in my system, so I'm not sure why it's saying they don't exist when they very clearly do).
 

paleoN

Wizard
Joined
Apr 22, 2012
Messages
1,403
I need to look into iperf more and I'm hoping the Ubuntu Live CD has iperf. That would put my on a level playing field and I wouldn't be using a VM on my Windoze machine.
So, no FreeNAS to FreeNAS testing? If iperf isn't on Ubuntu by default you should simply be able to install it.
 

joeschmuck

Old Man
Moderator
Joined
May 28, 2011
Messages
10,970
When I try to run the dd command from your post, I get: "dd: tmp.dat: Read-only file system".
I tried to mount it as RW, but it's telling me no such directory exists (I changed the directory to reflect actual folders in my system, so I'm not sure why it's saying they don't exist when they very clearly do).
My bad, the problem is you are not currently on a hard drive. "tmp.dat" is a file being written to the hard drive. You can do this one of two ways, I prefer changing the directory "cd" to a hard drive mount location. In my system it would be "cd /mnt/farm/ftp" and then run my command. This part of the testing only checks the internal speed of data transfer within your NAS and can take considerable time to run (easily over 5 minutes depending on drive setup). You really should perform something like Test 5 & 6 to get a feel for the speed.

Here is my result for the first dd command with my current system:
Code:
[root@freenas /mnt/farm/ftp]# dd if=/dev/zero of=tmp.dat bs=2048k count=50k   
51200+0 records in                                                             
51200+0 records out                                                           
107374182400 bytes transferred in 393.601444 secs (272799259 bytes/sec)
 

BrianDMG

Explorer
Joined
Jan 19, 2013
Messages
70
Alright, here are the results (I've only run them once each, both while the system was in use with MiniDLNA. I would prefer to take the average of 3, but it was taking a long time and the wife didn't want to give up watching her show):

Test 1
dd if=/dev/zero of=tmp.dat bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 1904.656304 secs (56374571 bytes/sec)

dd of=/dev/zero if=tmp.dat bs=2048k count=50k
51200+0 records in
51200+0 records out
107374182400 bytes transferred in 875.016263 secs (122711071 bytes/sec)

If it makes any difference, I issued these commands via the GUI shell. I couldn't tell you why the results are as slow as they are, but if you have any insight, I'd be curious to know.
 
Status
Not open for further replies.
Top