Re: [gentoo-user] Bad NFS performance
sounds like NFS packets are being dropped. Are you using UDP? Try using TCP... On Sun, 19 Jun 2005, Andreas Karlsson wrote: Hello all, I am getting quite sick of my self and NFS. I?ve read many posts at forums.gentoo.org, I?ve googled, I?ve tested and tested and tested... No help could be found. The problem is not unique as it seems. I have an r/w speed against the NFS-server going from ~200kb/sec to 2mb/sec. I can?t predict or find a cause for the strange speed changes. I have three computers set up: Aquarius is my main work station (AMD XP 3200+, 1 gig ram) Leo is my file server (AMD 1700+, 630 meg ram) Aries is for TV-out (AMD 1700+, 512 meg ram) All connected via a switched 100mbit network. All have 100mbit NICs. Netperf is telling me this: aquarius hermes # netperf -H leo TCP STREAM TEST to leo Recv SendSend Socket Socket Message Elapsed Size SizeSize Time Throughput bytes bytes bytessecs.10^6bits/sec 87380 16384 1638410.00 89.61 And when I do the same from/to all others the results are quite the same. So, no problem with the network. Next up NFS. I have now tried so many different combinations at my /etc/fstab that I don?t know what is up and down: It currently looks like this: aquarius hermes # cat /etc/fstab | grep nfs leo:/home/files /home/hermes/philez nfs rw,users,auto,hard,intr,tcp,retrans=5,rsize=1024,wsize=1024 I?ve tried rsize and wsize from 1024 up to 32768 (1 through 32 * 1024). I?ve tried it in combination with sync and async. Here I find async working best. The file server (Leo) has a 120 gig HD with DMA enabled: leo linux # hdparm -t /dev/hdc /dev/hdc: Timing buffered disk reads: 88 MB in 3.02 seconds = 29.15 MB/sec So there should not be a I/O related problem either. I also run Samba on it, and when I am in Windows, filetransfers are as expected on 100mbit network (haven?t even mesured them as they work fine). Can anyone please give me any hints or advices on how to proceed with this problem? Best regards, Andreas Karlsson -- Bryan Whitehead Email:[EMAIL PROTECTED]
Re: [gentoo-user] Bad NFS performance (SOLVED)
On Thursday 23 June 2005 00.37, Bryan Whitehead wrote: sounds like NFS packets are being dropped. Are you using UDP? No, I use TCP. The problem is now solved though. Strangely enough, when I removed a second switch and connected directly to the first switch I worked. So one of my switches is probably more or less broken. Best regards, Andreas Karlsson pgpEJrXglOBdT.pgp Description: PGP signature
Re: [gentoo-user] Bad NFS performance
On June 19, 2005 03:43 am, Andreas Karlsson wrote: On Sunday 19 June 2005 00.45, Scott Tiret wrote: On Sun, 2005-06-19 at 00:22 +0200, Andreas Karlsson wrote: Hello all, I am getting quite sick of my self and NFS. Ive read many posts at forums.gentoo.org, Ive googled, Ive tested and tested and tested... No help could be found. I had similar performance issues with NFS. Tried all the different NSF options. Turned out to be a simple networking problem. One of the hosts on my switched 100Mb/s network was actually running half-duplex. Try ifconfig to see if there are any errors on each of the network cards. Also try mii-diag to see what the auto-negotiation messages are saying. You can then set speed and duplex using mii-tool if you need to. You might also check the cables just to be sure. Thanks for the reply. I forgot to include that information. All cards are running at 100mbit, full duplex, with auto negotiation on, according to ethtool. And as I get good performance out of netperf I dont really see this as a problem on a harware level. Neither are ifconfig reporting any error-packets or droped packets. make sure you've got the async option set on the share. sync is now the default and it's slower to transfer. -- without law or compulsion, men would dwell in harmony - lau tzu, tao te ching: chapter xxxii __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com -- gentoo-user@gentoo.org mailing list
Re: [gentoo-user] Bad NFS performance
I'm getting pretty good performance with these settings: fstab on nfs mount client: naked:/mnt/share/ftp /mnt/nfs/ftp nfs defaults,auto,soft,rsize=8192,wsize=8192,bg,posix,nosuid,intr,tcp 0 0 exports on nfs mount host: /mnt/share/ftp 192.168.x.xxx(root_squash,rw,async,secure,nohide,anonuid=65534,anongid=65533) Note that the NFS host has nfs over tcp enabled, which (AFAIK) makes it much faster. Barring that, if you do the same thing with SMB, are the results similar? On Sun, Jun 19, 2005 at 12:22:50AM +0200, Andreas Karlsson wrote: Hello all, I am getting quite sick of my self and NFS. I?ve read many posts at forums.gentoo.org, I?ve googled, I?ve tested and tested and tested... No help could be found. The problem is not unique as it seems. I have an r/w speed against the NFS-server going from ~200kb/sec to 2mb/sec. I can?t predict or find a cause for the strange speed changes. I have three computers set up: Aquarius is my main work station (AMD XP 3200+, 1 gig ram) Leo is my file server (AMD 1700+, 630 meg ram) Aries is for TV-out (AMD 1700+, 512 meg ram) All connected via a switched 100mbit network. All have 100mbit NICs. Netperf is telling me this: aquarius hermes # netperf -H leo TCP STREAM TEST to leo Recv SendSend Socket Socket Message Elapsed Size SizeSize Time Throughput bytes bytes bytessecs.10^6bits/sec 87380 16384 1638410.00 89.61 And when I do the same from/to all others the results are quite the same. So, no problem with the network. Next up NFS. I have now tried so many different combinations at my /etc/fstab that I don?t know what is up and down: It currently looks like this: aquarius hermes # cat /etc/fstab | grep nfs leo:/home/files /home/hermes/philez nfs rw,users,auto,hard,intr,tcp,retrans=5,rsize=1024,wsize=1024 I?ve tried rsize and wsize from 1024 up to 32768 (1 through 32 * 1024). I?ve tried it in combination with sync and async. Here I find async working best. The file server (Leo) has a 120 gig HD with DMA enabled: leo linux # hdparm -t /dev/hdc /dev/hdc: Timing buffered disk reads: 88 MB in 3.02 seconds = 29.15 MB/sec So there should not be a I/O related problem either. I also run Samba on it, and when I am in Windows, filetransfers are as expected on 100mbit network (haven?t even mesured them as they work fine). Can anyone please give me any hints or advices on how to proceed with this problem? Best regards, Andreas Karlsson -- Alan [EMAIL PROTECTED] - http://arcterex.net Backups are for people who don't pray. -- big Mike pgpvK10WBGgZU.pgp Description: PGP signature
[gentoo-user] Bad NFS performance
Hello all, I am getting quite sick of my self and NFS. Ive read many posts at forums.gentoo.org, Ive googled, Ive tested and tested and tested... No help could be found. The problem is not unique as it seems. I have an r/w speed against the NFS-server going from ~200kb/sec to 2mb/sec. I cant predict or find a cause for the strange speed changes. I have three computers set up: Aquarius is my main work station (AMD XP 3200+, 1 gig ram) Leo is my file server (AMD 1700+, 630 meg ram) Aries is for TV-out (AMD 1700+, 512 meg ram) All connected via a switched 100mbit network. All have 100mbit NICs. Netperf is telling me this: aquarius hermes # netperf -H leo TCP STREAM TEST to leo Recv SendSend Socket Socket Message Elapsed Size SizeSize Time Throughput bytes bytes bytessecs.10^6bits/sec 87380 16384 1638410.00 89.61 And when I do the same from/to all others the results are quite the same. So, no problem with the network. Next up NFS. I have now tried so many different combinations at my /etc/fstab that I dont know what is up and down: It currently looks like this: aquarius hermes # cat /etc/fstab | grep nfs leo:/home/files /home/hermes/philez nfs rw,users,auto,hard,intr,tcp,retrans=5,rsize=1024,wsize=1024 Ive tried rsize and wsize from 1024 up to 32768 (1 through 32 * 1024). Ive tried it in combination with sync and async. Here I find async working best. The file server (Leo) has a 120 gig HD with DMA enabled: leo linux # hdparm -t /dev/hdc /dev/hdc: Timing buffered disk reads: 88 MB in 3.02 seconds = 29.15 MB/sec So there should not be a I/O related problem either. I also run Samba on it, and when I am in Windows, filetransfers are as expected on 100mbit network (havent even mesured them as they work fine). Can anyone please give me any hints or advices on how to proceed with this problem? Best regards, Andreas Karlsson pgprVgpW0Qkn7.pgp Description: PGP signature