I have a bottleneck in my home network, and I can't find it. Any ideas?
December 23, 2008 9:03 AM   Subscribe

I have a bottleneck in my home network, and I can't find it. Any ideas?

I've had a home fileserver for a while now, and have had speed problems with it from the start. For some reason the speed I get always tops out at around 10-15 Mb/s, when the network is capable of handling 100 Mb/s. My network is pretty simple, with a WRT54GL running Tomato doing the routing, and my fileserver and two other computers running on the network with Cat 6 cables all around. None of the computers are more than 6 feet away from the router. My speeds from both the computers to and from the fileserver are limited to 10-15 Mb/s. However, my speed to and from the two other computers on the network reach the 100 Mb/s threshold. (btw, I've been determining these speeds using this method). This problem has endured through two different computers, one a d201gly2 self built machine and a Power Mac G4 Gigabit Ethernet model I had lying around. This problem even still occurred on the d201gly2 based machine when using FreeNAS, Windows Server 2003 and 2008. I have also used a Cat 5e cable and a Cat 6 cable, but still the problem remains.

Up to this point I have done a lot of things, but not everything. All these different configurations have been running RAID 5 in either software or hardware, using a Highpoint 1740 card. Should I try a different RAID level or JBOD? Another constant has been the router. While I've been able to demonstrate that I can reach decent speeds with it, it's something else that has remained the same. Should I swap it out? Any help you guys could offer would be greatly appreciated.
posted by northernsoul to Computers & Internet (16 answers total) 1 user marked this as a favorite
 
I'd be switching out the router. I've personally had bad luck with Linksys and better luck with Netgear and D-Link. But a search of AskMe will find out a lot of general dissatisfaction with consumer grade routers. In any case it seems like the logical next step to try.
posted by 6550 at 9:37 AM on December 23, 2008


check that the connection is actually 100Mb and full duplex. You could be running at 10/half.
posted by roue at 9:39 AM on December 23, 2008


It may take some work, but if you can setup two of your machines for p2p networking (using a crossover cable) that will allow you test for speeds without the switch. If the speed goes up, you know it's a cabling or switch problem. If it stays the same you know the problem is likely with disk access or some other system specific issue.

I'd lean toward a switching issue. Gigbit switches and cards are cheap nowadays, maybe it's time to upgrade?
posted by wfrgms at 9:45 AM on December 23, 2008


Response by poster: Just buying a Gigabit switch was my next option. I like Tomato so I'd rather not just replace the router completely. I am also hesitant to keep throwing money at the problem, since this has already taken a bit more than I had originally budgeted.

roue, I'm not familiar with how to check that. How would I go about doing that?
posted by northernsoul at 9:57 AM on December 23, 2008


well if the router is a 10/100 the transfer rate of 10-15mb/s would be about right. thats the difference between megabits and megabytes. 100 megabits is equal to about 12.5 megabytes per second. if you want you can get a gigabit router and that will improve your speed to around 100 megabytes per second providing you also have a gig nic card. then your limiting factor can be your hard drive if you're running an older hdd or one of the newer wd green hard drives.
posted by no bueno at 10:07 AM on December 23, 2008


What notation is Mb/s? If youre saying megabytes per second then thats what 100mbps gets you.

The method you are using to determine speed doesnt seem to let you try different packet sizes. I suggest trying iperf and trying different size frames to see if it reports anything different.

Are you using the WAN port on your linksys for your LAN? If so its 10mbps port and tomato or dd-wrt cant change that. If anything is plugged into that port I would remove it.

Are any of the servers or client running anti-virus software or any type of malware scanner? if so, disable it. These apps kill network performance.

Should I swap it out?

Im sure you can borrow a plain jane 100mbps switch from someone or buy one from a store with a generous return policy.
posted by damn dirty ape at 10:45 AM on December 23, 2008


You're not going to get it. MS TCP stack is fucked up. The thing is that it ramps up, you can't push mass amounts down the pipe because it wasn't designed that way... It's designed to play nice with others... you want to make it fast and you're tweaking registry keys and buffers and timeouts. You have a limiter on your 'puter, it makes it play nice, trying for fast is hard work. You can tweak it, it'll make your life more hellish.
posted by zengargoyle at 10:47 AM on December 23, 2008


Bullshit. Every POS Windows machine Ive ever used, supported, or built has been able to max out the ethernet bottleneck. If youre getting 10mbps with a 100mbps card then you have a problem. Its not a windows limitation.
posted by damn dirty ape at 10:49 AM on December 23, 2008


Yeah I call BS on that anti-MS stuff. Maybe small variations, but off by 10 times? Doubtful.


I don't think it's the router either, because I'm sure you've tried various ports. It does sound like the media auto-detection could be defaulting to 10/half duplex, which would explain the very specific wall you're hitting.

You can find this in XP in the Properties window of the network connection, Configure the network card, under the Advanced tab.

Set it to be explicitly 100/full and see if it can still connect (it should). You should experience better speed.

One other thing: Tomato could be configured incorrectly. It might be worth backing up your Tomato config, flashing the router to factory, and trying your speed tests from there.

Also I assume by your notation you've been comparing bits to bits and bytes to bytes... that tends to throw people as well (network speeds are notated in bits, while most speeds calculated in software are represented in bytes, which will cause a difference of roughly 8 times what you would expect).
posted by teabag at 10:59 AM on December 23, 2008


As other folks have noted, make sure you're not confusing Megabits/second with Megabytes/second. pv, the tool you're using to measure, reports rates in megabytes / second. If you're seeing it report 10MB/s then you're getting 80+megabits/second, which is what you should expect on a good 100Mbps ethernet.

If you're truly seeing 10-15 megabits/second, then I'm confused. My #1 guess would be your ethernet is only running at 10Mbps, but then applications will be more like 8Mbps, you couldn't possibly be seeing 15. So if you're getting 15 megabits/second then things are running at 100Mbps and the problem is almost definitely bad network hardware or a terrible driver on one of the two computers.

Related, but I don't think it's your problem. I had trouble with a WRT54GL running Tomato getting 100Mbps out of my gigabit ethernet card. It kept going down to 10Mbps. You can see this quite clearly by looking at the blinking lights on the router's switch ports. I ended up getting a gigabit ethernet switch (I like Netgear) and now get 100 megabit on my home network with no problem.
posted by Nelson at 11:36 AM on December 23, 2008


If you decide to get a switch, I would recommend this one. It works great, is gigabit for when you upgrade, and runs very cool (and saves energy) because it gives each port only the power it needs for the length of the attached cable. I regularly transfer large video files from one machine to another in around a minute.
posted by kindall at 11:59 AM on December 23, 2008


northernsoul, I'm afraid I don't know much about Windows machines, but on Mac boxen, if you get into the Terminal and type ifconfig, you will get a lot of information about the connectivity and protocols of the different sockets you are running, whether they are up or down, what speed and duplex they're running, all that stuff. You can get an abbreviated version of this info from the Apple System Profiler in the Utilities folder. This can be useful for troubleshooting these types of issues.

There is probably an analogous utilty hiding somewhere on XP and Vista boxen, but I would not know where to start looking for it.
posted by ikkyu2 at 2:54 PM on December 23, 2008


Just throwing this in, though it will probably not be the cause of your problem: just recently I've become aware that telling Windows to force a Marvell Yukon gigabit-capable NIC to operate at 100mpbs full duplex instead of letting it autonegotiate the speed will cause lots of packet loss on that port, to the point where it's not really usable for Windows networking, even when the other end of the cable is plugged into a 100mbit or gigabit switch.

But back to your problem: Things you've changed without making the problem go away:
* file server computer including the NIC (d201gly2, Power Mac G4)
* file server operating system (Windows, BSD/FreeNAS)
* cables

Things you haven't changed:
* router/switch

Things I don't know if you've changed:
* which router/switch connectors are assigned to which computers

Personally, I'd be looking sideways at the router.
posted by flabdablet at 3:10 PM on December 23, 2008


Make sure tomato isn't doing any QOS or port/mac based throttling.

Other than that, you'll have to see what the server machine is doing when you are running the test. Think about the PCI bus being an issue- the NIC and the RAID card are probably both trying to use it (and who know what other devices) and bringing throughput to its knees. Maybe the RAID card doesn't play nice?
posted by gjc at 5:40 PM on December 23, 2008


Oh, and how are you running the linux commands on the windows box(s)? Whatever method used is bound to cause issues. Including the possibility that the command interpreter is limited to 10/mb/s...
posted by gjc at 5:48 PM on December 23, 2008


After running that test, I'm going to go with "you're counting bytes instead of bits, and there's nothing wrong with your setup."
posted by oaf at 7:28 AM on December 24, 2008


« Older AAAGH!   |   This is such a crazy long shot Newer »
This thread is closed to new comments.