Oh, Sky Cake Windows. You really are a new toy every day, aren't you?
Readers with unusually long memories may remember that I shamelessly begged for money to buy a new computer. Against all reason, you actually gave me enough to make that possible, just before the end of the last financial year. Said new computer, replete with overclocked Core i7 920 CPU and 6Gb of RAM, has been happily buzzing away next to my desk ever since.
I'm not actually using the new computer yet, though, because I will not permit myself to start screwing around playing Fallout and GTA and such on it until I have actually finished writing a big review about it, like unto the piece I wrote about the Athlon X2 box in 2006, and the other piece I wrote about the Pentium 4 box that preceded it in 2003.
But every time I get back to working on that big review, the PC bang-per-buck goalposts have shifted again. There is, for example, not really much reason for most people to get an LGA 1366 Core i7 machine any more, now that functionally-no-slower, yet cheaper, LGA 1156 CPUs are available. And don't even start me on the graphics-card scene.
So this has turned into the longest PC-to-PC migration project in history, with the new machine being languidly updated with data and applications. It's on all day, but only actually running a BitTorrent and distributed.net client. (I think you can spot the moment in my stats when the new box came on line. Feel free to mess up the numbers by ascribing your own distributed.net work to dan@dansdata.com, too.)
So anyway, the new computer's running Windows Vista SP2 (the 64-bit version, so I'll be able to use all of the 6Gb of memory), and it behaved itself perfectly for weeks on end. As you'd expect it to, of course; Vista was something of an adventure in frustration when it was freshly birthed, all shiny and glistening, but the two service packs have burned away the more impudent of its tentacles.
But then, just the other day, the Vista box decided to stop moving data over its gigabit-Ethernet link to my old computer, the one I'm still using, at the tens of megabytes per second to which I'd become accustomed.
Instead, it's decided to send data at, oh, maybe half a megabyte per second. 1.5Mb/s, tops. Often quite a lot less.
That screen grab is of a transfer from the Vista machine to the XP machine, initiated and screenshotted at the Vista end. But speeds are the same if I start a transfer from the XP end.
It sped up to about 200 kilobytes per second after a few minutes. Sometimes, at random moments, it actually managed to sustain a whole couple of megabytes per second for a while. Whoopee.
Copying between all the other devices on the network works exactly as fast as it always did. The Vista box copies files between its own drives very quickly. The laptops get full bandwidth from their wireless adapters, the Vista box copies to the little Thecus N299 at its usual roughly 8Mb/s, and copying from the XP box to anything else on the network is also fine. And, get this, copying from the XP box to the Vista box is fine, too. Full gigabit speed. So this is a one-way problem.
And it's specific to the (Realtek) network adapter on the Vista box's Asus P6T motherboard. When I unplugged the Ethernet cable and plugged a USB wireless adapter into the Vista machine, I got full wireless bandwidth from Vista, via the access point and its own Ethernet hookup, to my XP PC. I presume a PCI Ethernet card or USB Ethernet adapter would work fine, too - though I wouldn't be at all surprised if the slow-transfer disease spread to the new adapter in due course.
I've plugged the Ethernet cable back into the Vista machine's built-in adapter for simplicity, now. Since the BitTorrent client is on the Vista box, this means that if I download something big on the Vista machine and want to move it to the XP one, I can either start it copying long before I want it, or plug a thumb drive into the Vista box, copy the file (at a perfectly normal speed) onto that, then plug the thumb drive into the XP machine.
This problem - or something very like it - was all the rage among Vista's early adopters back in 2007. I think the 2007 version of the problem usually had to with a well-meaning feature in Vista which is supposed to reserve network bandwidth for streaming multimedia content, so if you're watching an HD movie or something over a (suitably speedy) network from a Vista computer, you'll never have any frame-dropping or glitches when seeking, because any other file transfers from that computer will be heavily throttled even when they don't need to be.
This feature apparently often went haywire, especially in the original version of Vista. It either decided to operate all the time whether you were playing video or not, or it operated when the local user of the Vista computer was just playing music, or something, while someone else tried to get a file from his computer over the network. I think there was some kind of Copy Control Crap involvement here, too, but don't quote me.
This was meant to be fixed in SP1, and by all accounts a lot of it was. Vista Service Pack 2 has been out for some time now, and that's what my new computer is running. And as I said, for weeks on end, everything worked fine. I could play even HD movies from the Vista box over the network, A-OK.
Because this problem has such a long history, it's somewhat challenging to dig up information about fixing it on Vista SP2, as opposed to SP1 or the original extra-special Oh Dear God Why Did I Buy Vista v1.0 Edition. An inexpertly-crafted search string will thus turn up tons of people complaining about it back in 2007. The water is further muddied by different versions of the problem, in which copies from Vista to, say, Windows Server 2003 work OK, but copying stuff the other way is very slow and may even time out and die entirely. I don't think my problem is related to those ones, but who frickin' knows.
I have tried many things to fix this problem.
[UPDATE: In the original version of this post I forgot to mention that yes, I'm using a full-permissions administrator account, and yes, Vista's firewall is turned off.]
First up, I tried using a different copying program (like the aforementioned TeraCopy, or Vista's own Robocopy). No good.
I tried opening a DOS prompt (with admin permissions) and typing the voodoo chant "netsh int tcp set global autotuninglevel=disable"
. No good.
I tried Microsoft's automatic "Fix It" doodad for changing this same setting. No good.
I noticed that the Vista PC's hard-drive light is locked on for a couple of solid minutes after startup, even if I close all apps that could be expected to hit the drive. I don't remember whether it did this before the transfer problem. Perhaps it's SuperFetch-related. While I was fiddling with this, I completely disabled Windows Search. No good. Didn't even prevent the drive-flogging on startup.
(Good old lsass.exe was totting up I/O reads and writes at a great rate. I'm unconvinced that it had much to do with the startup disk-flogging, though, since it kept on reading and writing after the drive light had returned to normal occasional flashing.)
I've got Nero installed on the Vista machine; that installs some pointless services that can also hit the disk, so I killed them along with indexing. I also disabled Nero's system-startup tasks using MSConfig. No good.
I power-cycled the cheap and cheerful gigabit switch. No good.
I usually have a VNC view of the new computer's desktop open. I closed that. No good.
(VNC itself is subject to the slow-transfer problem; it updates very noticeably slower now, and of course becomes even more painful if I ignore the new limited bandwidth and force it to a high-bandwidth connection mode, like the "LAN" setting in the UltraVNC viewer.)
I ventured into the registry, and tried setting NetworkThrottlingIndex to FFFFFFFF. Then rebooted. No good.
In a moment of mad optimism, I tried telling Vista to "diagnose and repair" the network connection. It told me I needed to "turn on TCP performance improving settings", so I did. No good.
I turned off Quality of Service for the XP machine's network adapter. No good.
I turned off the same QoS Packet Scheduler and a couple of Link-Layer Topology Discovery doodads on the Vista box's network-adapter properties. No good.
I tried mapping a drive. No good.
I went on a rampage through Task Manager, killing every task that wasn't obviously necessary for Vista's continued operation. AnyDVD, audiodg, Daemon Tools, GoogleCrashHandler, jusched, nTuneService, PunkBuster, UpdateCenterService, Real Temp, PresentationFontCache, nvSCPAPISvr, MSASCui, two copies of nvvsvc.exe, Vuze and the VNC server all bit the dust.
No good.
But then there was SLsvc.exe, a Copy Control Crap process if ever I saw one. I killed it, and... No good.
I fiddled with "Remote Differential Compression". Windows said "Please wait while the features are configured. This might take several minutes", and for once it was not joking. It sat there for quite a while. But then it finished! No good!
I read through this page looking for things I hadn't yet tried. The only new one I found was disabling "Windows Meeting Space". So I did that. No good.
I said, "hang on a minute - why not just connect the XP and Vista boxes with a FireWire cable? That's fast!"
So I did. And although the XP machine was perfectly willing, it didn't work at all, because Microsoft has removed FireWire networking from Windows, as of Vista.
I noticed that Windows Update had a new driver for the motherboard's network adapter, which I hadn't installed with the other updates. So I installed that. No good.
I tried disabling "Large Send Offload" in the Vista machine's network-adapter properties. I even disabled the IPv6 one as well as the two IPv4 ones. No good.
While I was there, I tried disabling the IPv4 and IPv6 versions of TCP and UDP Checksum Offload, and an IPv4 Checksum Offload too. Each of them can be enabled for receiving, transmitting, both, or neither; I fully disabled all of 'em. No good.
I went to Device Manager and uninstalled the network adapter - and selected the "Delete driver software" option - then rebooted so it'd be redetected. No good.
Then I smote my forehead mightily, and tried a new Ethernet cable. I would actually have been slightly irritated if that had worked. It didn't. Actually, it made XP-box-to-Vista-box copies slow, just like Vista-box-to-XP-box ones. Both cables have all four pairs connected - well, unless there's a break in the middle somewhere. I cannot escape the feeling that this is trying to tell me something, but I'm too tired to figure out what it is.
I haven't yet tried starting the Vista box in Safe Mode with Networking, as this page suggests. I haven't tried connecting the two computers with a crossover cable, either. I also haven't yet tried just officially declaring the migration to be complete and starting to use the new box as my main computer.
But dammit, I want to fix this. I've gone too far to turn back now.
Perhaps there's something obvious that I'm missing, here. If any of the three people who've managed to read to the end of this post have any suggestions, I'm all ears.
19 October 2009 at 3:32 am
This looks a lot like an intermittent problem I've seen on servers with Broadcom GigE adapters - have you tried fixing the adapter speed to 1000Mbps Full duplex rather than leaving it at Auto. You are supposed to leave it on Auto but quite a few adapter\switch combos have a problem with that. Also try [a] changing the switch port, [b] change the switch all together, [c] if all else fails does a directly connected ethernet link also suffer from the same problem? and [d] Get a decent GigE adapter - a nice Intel Pro1000 should do the trick and try that.
The fundamental behavior that you are seeing (things are OK over wireless more or less) indicates to me that this is a low level problem at the ethernet layer and not higher up the stack where you have spent most of your time debuggging.
19 October 2009 at 3:43 am
You could boot Linux Live-CDs all over the place and test network speed with those. This would at least allow you to figure out if it's a software or hardware problem...
19 October 2009 at 6:05 am
I had/have a similar issue under Vista and Win7 RC1, only it was copying to a ReadyNAS box that suddenly slowed to a crawl. Reads were fine, copying to/from other machines worked without problems. While it's a ridiculous way to try and solve the issue, try disabling the print spooler service on the Vista side.
I can't even begin to fathom why that would cause problems with network transfer, and yet... it solved it in my case. (Of course printing becomes a pain, so it's more of a band-aid.) I didn't have a wireless connection to test against, so it's certainly possible that this is just another unrelated oddity.
19 October 2009 at 6:21 am
I had a similar problem that did end up being the cable. Given that the first cable swap you did caused an additional problem, I would try a third cable just to eliminate it as a possible source.
I also like the linux live cd idea to separate the hardware from the software.
19 October 2009 at 6:30 am
You don't mention looking in the Reliability and Performance Monitoring thingy to see if you're actually getting full disk or network IO usage according to windows. (Yes, unlikely, but when hunting Grues you need to look in dark places)
I would also run up iperf on both boxes and see what raw TCP speeds are like.
I was able to achieve just a bit over 900Mbit throughput between a Win7 Laptop (Core2Duo, 6GB RAM) and a Win2008 Desktop (Core2Quad, 8GB RAM). It took a bit of magic incantation to get it to that level though - default settings on iperf hovered somewhere around 200Mbit or so.
19 October 2009 at 6:48 am
I'd look at the switch - HP now make some outstandingly good yet reasonably prices gigE switches - the 1810G series if you want web management:
http://www.procurve.com/products/switches/HP_ProCurve_1810G_Switch_Series/overview.htm
and the 1400 series which are virtually identical, except they lack web management and therefore are cheaper.
http://www.procurve.com/products/switches/HP_ProCurve_Switch_1400_Series/overview.htm
If you went with one of the 1810 series switches, and have two Ethernet ports on your new beast, you can use LACP trunking to the switch to aggregate the bandwidth of the two network ports. Not that this will help if each port is only pulling 200kB/s but configuring the trunk may just fix whatever the hell it is that's wrong with the network connection in the first place
19 October 2009 at 7:30 am
I don't think I'd even bother fixing it with the most excellent Windows 7 right around the corner.
19 October 2009 at 7:49 am
I had a similar problem.
Run a ping -t and see if any packets actually get lost.
In my case, it was that the PC was sending jumbo frames and they were being dropped somewhere.
Would also explain the one way thing.
Turned off jumbo frames in the NIC settings and, presto chango.
Now you've changed so much that you should just reinstall - with Vista SP3, aka Windows 7.
Been running 7 for months now and no issues.
Mind you, ran Vista for years and had negligible issues. I think it's to do with the way you hold your mouth when you boot.
19 October 2009 at 8:00 am
Hi Dan
I'm feeling your pain here, I had all sorts of network problems with a shiny new Asus M2N32 SLI Deluxe board and Vista 1.0 (yes, pity me now). One of my problems was that I could browse the internet fine, as long as I didn't want to view a Microsoft site. Any microsoft site. Not just the "microsoft.com" domain, but anything related to them. I found a couple of other people asking the same questions as me, but there were never any answers to that particular question that I could find. I don't think any of the other readers actually believed those of us who had the problem...
In the end, I disabled both on-board NICs in the BIOS and installed a $15 PCI network card. That fixed it, simple as that. When I recently upgraded the processor to one of the newer tricore AMD models, I reset the BIOS back to its default settings and my network stopped working again. As soon as I disabled the on-board NICs again, everything was fine.
My solution isn't a "fix", more a workaround, but it suited me. It might keep you going until you find a more elegant solution.
19 October 2009 at 8:42 am
I agree with Hubris, your not poking your tongue out enough and its failing POST.... :)
You have way too much patience, I'd have nuked it up to "SP3" at the first sign of trouble...
p.s. the captcha just gave me an acid flashback to the 90's!!!
19 October 2009 at 8:46 am
I ran into this exact same issue with my Win XP and Vista box. Copying one way worked, the other way was exceptionally slow and would time out if the file was anything bigger than 2 megs. Tried many of the same fixes you did, could not find a solution at all.
My end solution? Convert my entire network to Win7.
19 October 2009 at 9:44 am
Time to get a Mac, Dan.
You'll even be forgiven* you if you run a Hackintosh.
*forgiveness may not apply in all jurisdictions/theologies.
19 October 2009 at 10:19 am
I wonder if it's a problem with the XP machine?
Of course, I'd still start with a third cable, and live-cds everywhere first before messing around with that.
19 October 2009 at 10:19 am
Nothing constructive to add, but I'd wondered where you'd been for the last week or so.
19 October 2009 at 11:11 am
I've got no idea about your network problem (except the usual cable-jiggling, tongue-poking, 'try Linux', etc...but I think the suggestion of moving to Mac was a bit harsh).
I just think it's cool that you're still dedicated to Distributed.net. Long after all the cool kids have moved to Folding@Home and such, it's good to see there's a core of diehards like us still dedicated to one of the original distributed computing projects. I've had a cow client running on every computer I could possibly lay my hands on, and get permission for, since 1997. Moo!
19 October 2009 at 1:43 pm
At any stage when you were updating Vista did a newer network adapter driver get installed?
The reason I ask is at one stage I downloaded a newer driver for the network adapter on my machine at the behest of the Windows Update service and I suffered from a range of different speed and drop-out issues.
The only fix for the problem was to uninstall the adapter and reinstall it using the old drivers. Any version of the drivers newer than the ones I started with caused these issues.
It took me a while to get it to work because after uninstalling the adapter I went and found the newest drivers for it for the re-installation/re-detection.
19 October 2009 at 1:57 pm
Seriously Dan, with the Excellent Windows 7 just around the corner, i strongly advise that you cease all activities related to improving your "vista experience" as Microsoft would put it, and wait till October the 22nd, when Windows 7 will be publicly available.
19 October 2009 at 2:16 pm
RE #16 - yeah, I had a similar problem with audio drivers on my previous box - I'd get horrible graphical corruption (missing textures, corrupted geometry vertices, and the like) when playing 3D games, and tried a bazillion different video drivers to try and fix it. Then I realised the stuttering in the sound when the graphical corruption happened might be a sign, so I tried a different sound card driver. Not the one Windows update suggested, not the one from my mobo manufacturer, but the one from the audio chipset manufacturer... completely solved that problem!
Some good suggestions here for things to try re the networking problem - I've also seen boxes that wouldn't go above 1-2Mbps unless you forced it to use 100Mbps, which then worked perfectly.
19 October 2009 at 3:08 pm
Bad port on the switch maybe? I've seen countless routers and switches with given percentages of their ports taped over with colourful things like "Bad!" "Do Not Use!" and the ever-popular "NFG" written on the tape... Just a thought.
19 October 2009 at 5:56 pm
A quick note about crossover-cables: They don't work for Gigabit. As you say, Gigabit requires all four cable pairs, and a crosslink cable only has two pairs (the other two go the other way), so it'll automatically down-grade to 100Mbit. And its autosensing, so you wouldn't need a cross-link cable in the first place, the pins would be switched in the adapter.
One thing you haven't mentioned you did is uninstalling the network card, rebooting, and have Windows reinstall it. I'm network support in my apartment building, and it's amazing how many problems get solved by that act.
And as mentioned above, plugging in a good PCI or PCIe network card is an interesting experiment as well (and will, with the right adapter, probably be faster if it works. I recommend Intel PCIe cards.)
I also second all the above comments about Windows 7. I just made a completely pain-free upgrade from Win2k (!) to Windows 7 64bit, and am very satisfied.
Good luck.
19 October 2009 at 6:03 pm
Only suggestion I have is do what usually seems to work for me when my computer is misbehaving. Swear at it, turn it off, come back in the morning. Usually working then.:P
I'm planning on upgrading to an i5 soon. woohoo! According to benchmarks on the internet I expect the CPu to be a bit over 5 times as fast and video 15 times as fast as the computer I've had for about 4 years.:)
19 October 2009 at 7:35 pm
I would like to re-emphasize this point which was already made by "mak_elblotto":
I have and have built several machines with Realtek GbE cards running various flavors of Vista 64-bit and I have had to disable Windows Update from installing the updated drivers for the damn things. The drivers that come from Windows Update are just plain wonky and may be the source of your frustration.
If you're using a Realtek 8111/8112 Family PCI-E GbE NIC, find and install version 6.217.226.2009 (dated 2/26/2009). If your motherboard has another vendor's GbE NIC, I recommend rolling back the driver or installing the one that originally came on the board's included driver disc.
Additionally, check your Windows Update history (Start->Windows update->View Update History) to see if such updates were installed within a few days of when you first started noticing this problem.
19 October 2009 at 7:42 pm
Have you tried removing recently installed patches? As this problem suddenly appeared out of nowhere, it's possible a patch caused it.
20 October 2009 at 12:37 am
Dammit, I read all the way to the end hoping to learn some new snazzy incantation for unborking network problems.
I agree with the earlier commenter that suggested booting with some kind of LiveCD to ensure it's not a hardware problem. I've had a switch go whonkers where it would happily use 7 of its 8 ports, but never 8 of its 8 ports. And the 8th port would intermittently go slow, and only with one specific computer. Move swap the wonky computer with another port, still slow. Unplug one of the other ports, it would work.
Drove me mad. And these were Macs, too.
20 October 2009 at 3:02 am
For the uninitiated- a crossover cable doesn't work like that. All a crossover cable does is switch the pairs around. For normal 2 pair ethernet, the same two pairs are only used.
What probably happens is that the other two pairs aren't crossed over in a "normal" crossover cable.
As for the problem- could it be some kind of netbios issue? The master browser disappears, or something like that, and the packets have to travel to your router to realize where they need to go??
20 October 2009 at 5:24 am
I'm not very Vista savvy but the print spooler idea rings a bell from a previous trouble shooting session. I think it was bringing the PC to its knees. I am just amazed that anyone can get windows to do any sort of networking other than connecting to a router, connect to another pc, dream on, use the file and setting transfer wizard, you must be joking! Definitely try yet another cable, maybe have a look at the socket to see if the pins are where they should be and not bent or unspringy. I don't know if Vista uses a system file to buffer transfers, maybe that has reached its limit and needs purging. Could there be a loop where data is being written back to the originating drive? I am guessing you have nailed system restore to the wall in case that is interfering? If it helps, I have a wireless card that just keeps dropping its connection and many an unhappy hour has been spent trying reinstalling software and enabling this and disabling that. My favourite network quirk is when two computers insist on using the same IP address and neither will release it.
20 October 2009 at 7:12 am
I had a similar problem, and in my case it was the offloading of the checksum. Apparently this is an issue with broadcom based network cards in Windows 2003 server & Vista. It seems like you've tried this though so what I would also try is giving your card a new mac address. This has worked for me before on similar oddd network problems.
Whilst I have to agree with most other posters about win 7 being great - one word of warning one one of my pc's I just can't get it to use hardwired connection to my router either via onboard card or another pci card, it just wont get a dhcp ip address. Consequently, I am now connected with a £15 wireless usb adapter to my router which is about a foot away.
If I could get all the apps I want ,I would get a frigged version of OS X and use that instead !
20 October 2009 at 1:43 pm
I too have had poor performance from upgraded Realtek drivers. Use driver rollback in Device Manager to return to the original driver.
It's got to the point that I now buy an Intel NIC for everything that I want to have running well (so some of my machines have 3 of them).
Also, I don't think I need to tell you (but other readers may not know) - network speed auto-detection can fail dismally if you hard-set one end only. You need to set both ends the same or it falls back to 10M half on the automatic end.
20 October 2009 at 4:15 pm
I'm with the others who suspect some kind of hardware issue. This appears to be the only thing the Cable Clue could possibly be telling you. I'm also with helvick in suspecting some sort of auto-negotiation issue. A top speed of 1.5megabyte/sec ties in quite nicely with good old 10 megabit per second ethernet, note. This could tie in nicely with Red October's idea of a dodgy port on the switch, too.
As I'm a Linux type I'd be running mii-tool or ethtool (Linux Rule #1: one small command line utility is *never enough*) to see what protocol had been negotiated, like so:
[root@htpc adamw]# ethtool eth1
(yadda yadda)
Advertised auto-negotiation: Yes
Speed: 1000Mb/s
Duplex: Full
(yadda yadda)
So I'm running at a full-on blazing gigabit, just as I'd expect for that system. If I were seeing 10Mb/s, that would be bad.
I presume there's some similar command / utility available for Windows, if you look for one. Some switches/routers do tell you what speed each connected device is running at via the status LEDs, too (my 'Linksys' router uses green for gigabit and blue for 100 megabit).
20 October 2009 at 4:58 pm
I have no idea what the problem is, but it sounds like a good excuse to buy yourself a $30 Marvell NIC and move to Windows 7. I've pushed this one to better than 90 MB/s.
http://www.newegg.com/Product/Product.aspx?Item=N82E16833166015
20 October 2009 at 5:48 pm
I have to disagree with the others in that I don't think it's a hardware problem. (I suppose you've already looked at the LEDs on the switch, and ensured that it autonegotiates to 1000Base-T.)
What I would have done (and have done in similar cases) is to simply look at what's going on. Do you have any experience with Ethernet analysers, such as wireshark (used to be called Ethereal)? It can be quite revealing to see what's really going on on your network. It wouldn't surprise me if you'll find that a large fraction of the available bandwidth is taken up by some service that shouldn't really be there, or by one computer re-sending some kind of request that it doesn't get an answer to.
I have had cases where similar network slowdowns were caused by inconsistent ARP tables, IP address duplication and the like, and unless you can actually see what's going on, you're just fumbling in the dark.
(And for all the OS-bashing fanboys out there - it's not an excuse. Wireshark runs just as well under OS-X and Linux as well as most flavours of Windows.)
20 October 2009 at 6:29 pm
My bet would also be a wacky network adapter driver. I have seen it all, on XP as well as Vista. Serves me right for buying Asus mobos with nVidia chipsets - I promise I'll never do it again. I had a problem of Xp dropping TCP connections randomly (effectively preventing driver update downloads;-)) on an nVidia network adapter, and also Vista misbehaving on a good old Realtek 8129. The solution was to use the one with the other and vice versa. Then, in about a year's time, XP and Vista service packs came (SP3 and SP1 respectively), and all problems disappeared auto-magically. I suspect a "silent" driver update or something in the TCP stack on both systems.
One more thing: what firewall software are you using? I found out that my two windows installs are being increasingly picky with the right versions recently. I had to ditch ZoneAlarm on Vista (it begun to drop internet connection altogether after about an hour of a reboot), and after that Outpost personal (for a very similar behaviour in a shorter amount of time;-). Ended up using the built in Vista firewall (shock! horror!!!), and an older version of ZA on XP. But then I'm sitting behind a router, so not much actually gets to me from the outside. If I were to say anything about it, I'd say that the firewall industry has gone a bit downhill recently, more Vista-like in quality. If anyone knows of any decent personal (preferably free) firewall that works on Vista, let me know please.
Last thing, which was also suggested, is to use wireshark to sniff the problem out (but if the traffic in the other direction runs at 1Gbit/s, I don't see how that could be the case).
21 October 2009 at 12:13 am
Wireshark. Use it to find out exactly what is happening on the network.
My bet for the pool is *not* a hardware issue, despite how much Realtek chips suck. I think it's a software issue here.
Furthermore, I believe it's Vista moving one byte at a time per SMB packet. I've seen that happen and it results in VERY slow network speed. I haven't bothered with a fix because it was a Windows work machine and I use Linux at home. I tease the IT guy with "Linux just works" every time Vista breaks (fairly regularly).
21 October 2009 at 4:46 am
+1 for wireshark. Look at the packets, see if they're going where they're supposed to go, whether they're acknowledged or that they're retransmitted without end, and also their size. Definitely look at both ends.
Also +1 for the Linux LiveCD. When you've booted that one, use something like ftp or rsync between the machines to see if that can reach acceptable performance levels, not something such as SMB that's bits wrapped in tinfoil wrapped in tupperware boxes wrapped in clingfilm wrapped in seaweed wrapped in used condoms, then stuffed in TCP packets.
It _is_, in all probability, something pretty low, as you've shown with the wireless adapter. That's drivers or hardware; the IP stack and system performance look to be quite OK. "Ditch the RealTek" _is_ a bit of a kneejerk reaction, but not totally so.
BTW, did something change elsewhere on the network at the time the problem started? And are the IP addresses DHCP or solidly nailed down? What's arp saying?
21 October 2009 at 11:22 am
Dan, just so you know, I too experienced this problem occasionally. I had Vista SP2 on my machine with broadcom gigabit (but connected at 100mbit, because the gigabit D-Link switch I bought REFUSES to move packets through at anything faster than what the slowest link connected is, so I junked it and am just connected directly through my WRT54G with DD-WRT). I'd see transfers through Samba to my Linux fileserver running at the full 100 mbit, then suddenly - a unidirectional drop to 200KB/sec or so. I could upload files to the fileserver at full bore, but downloading was a no-go and it was very frustrating. Downloads from the internet, frustratingly, are faster when this is happening. However, I didn't have your problem of it - a reboot would fix it, and the problem was much more intermittent. What someone said about TCP checksum offloading sounds about right - I'd turn it off and see what happens. For reference, I've an Asus P5Q Pro motherboard (speaking of CPU socket obsolescence - I bought it in January when LGA1366 was brand spanking new and was for "high end" chips only, then Intel promptly released 1156 and cheerfully announced "We're EOLing LGA775 and the Core2 branding, upgrade your hardware kthx!" I was not pleased.)
In the end, I ended up dusting off and nuking the site from orbit to install "Vista SP3", so perhaps as a last resort you might try that. The drivers are now included in the OS, I seem to recall having to install them from CD last time. I haven't seen the problem yet but then I've only been using it for a few days. I will report in if I see it here.
21 October 2009 at 11:26 am
Scratch that - I THOUGHT my gigabit NIC was Broadcom, but it's Atheros. So I guess that makes at least three manufacturers named in the comments, and what I said about TCP checksumming is probably wrong.
Firefox says I should be spelling Atheros Atherosclerosis. Maybe THAT explains it - clogged arteries, or, clogged tubes.
21 October 2009 at 10:45 pm
On an altogether different subject:
Is it a coincidence that Dan's piece on flying robots comes out on the same day as xkcd comments on the same thing?
22 October 2009 at 4:04 pm
Unrelated to the topic sorry, but do you know http://www.dansdata.com is blocked to a whole range of IP addresses and has been for some time?
http://forums.whirlpool.net.au/forum-replies-archive.cfm/1281659.html#
Emailed and twittered you about it but didn't hear back, and apparently neither did the whirlpool guys...
24 October 2009 at 2:45 pm
Or is there something with the harddisk...
I've had a different problem ever since I tested out Server2008, and even Server2008R2.
When copying files more than 100+ mb ie movies and such... the entire system will FREEZE for up to 1min. From HD to HD. From partition to partition. From a thumbdrive to HD. From HD to thumbdrive.
The system FREEZES. No cursor movement. Video stops. But audio begins to stutter like a fast skipping record.
A min later, everything will pop back to normal.
Can't find a solution to this. Makes me think that there's a serious issue somewhere with MS writing/reading the harddisk. Never happened when I was using XP with the same set of hardware.
25 October 2009 at 2:15 am
Total shot in the dark here. Maybe a waste of characters: Do you have Groove Folder Synchronization installed? It slows a lot of other crap down.
26 October 2009 at 3:05 am
If you haven't done a format and clean reinstall of the OS yet, you can add that to the list of things to try. Older versions of Windows (especially the 9x product line) needed this with much greater frequency, but I wouldn't entirely rule it out on newer systems.
Of course, any time you're doing a clean OS install, you always ask yourself, "Assuming this does solve the problem, is this the operating system I want to have when the install is all done, or is this a good chance to change it since I'm installing anyway?"
26 October 2009 at 4:34 am
I'd also like to place my vote on a crazy network adapter driver update delivered to you by our good friends at Windows Update.
I updated my Realtek driver from Windows Update once. Never again. The damned thing refused to work at all after the update, and until I rolled the driver back. Crazy shit.
Go have a look in your Device Manager. I bet you'll fix the problem with a simple rollback.
29 October 2009 at 4:04 am
At this point I think you might want to try one of my favorite last ditch solutions: Holy canned air, an old priest, and a young priest.
2 November 2009 at 5:15 pm
this story needs closure!!!
3 November 2009 at 4:15 am
Oh yes, Windows Update and drivers.
Recently it decided the one I was using for my somewhat aging but still perfectly functional ATI Radeon 7500 was past its use-by date, and offered me a Spiffy New More Betterder one to replace it.
After the obligatory reboot the screen came up in 640 x 480, at 4 colours. This, I considered, was somewhat suboptimal, especially since the box in question is mainly used for viewing movies. The screen resolution panel did offer all resolutions up to 1280 x 1024, and colour depth up to a bajillion hues, but the result of moving the slider and selecting "OK" was ... disappointing. Any effects were conspicuously absent. Moving it was as futile as banging a dead parrot on the counter in order to revive it. The card was incurably stuck at 640 x 480, and even copious cursing did not persuade it to behave otherwise.
So I invoked the Wrath of Driver Rollback, and one more reboot later all was well again.
3 November 2009 at 6:53 am
@unfunk - I agree.
You ask us for suggestions, we give some, and then...
nothing.
It's unlikely that you've made _no_ progress on this, given it's your new shiny machine and all, so I conclude that you're treating your readers somewhat shabbily.
Unless you've been unceremoniously replaced by a blonde female blogger because of some combination of "mass appeal", your poor work ethic and rising hubris...
4 November 2009 at 4:09 pm
Had a very similar problem, spent about a month pulling my hair trying to figure out wtf was going on. Turned out to be an intermittently bad Netgear switch (the cheap and cheerful variety, Model GS605 if you're interested). If the damn thing had just failed it would have made things much easier, but instead it would randomly slow down some connections to a crawl while everything still indicated gigabit life was good. One of the most frustrating problems I've ever had to troubleshoot...
22 December 2009 at 9:57 pm
I had this problem tonight! But it was when copying from an OS X share via Vista. Didn't matter at what point on the network I put it, it was incredibly slow and often just dropped out. But then when I initiated the transfer on the OS X end, it worked beautifully and saturated the link between them. What on earth is going on?