What version of Linux? If its 7.3, then do a search on NOW for redhat 7.3+NFS. Its a known bug,a nd the RH Kernel needs upgrading to fix it properly.
Simon
-----Original Message----- From: Stefan Funke [mailto:bundy@arcor-ip.de] Sent: 17 June 2003 10:15 To: toasters@mathworks.com Subject: netdiag / linux nfs over tcp
Hi all!
Is someone having a lot of linux clients mounted a toaster via nfs over tcp? I have a mixed environment of linux, bsd and sunos clients. When I run "netdiag" here the toaster finds all linux hosts not to perform like they should:
Average size of NFS TCP packets received from host: foo is 461. This is less than the MTU (1500 bytes) of the interface involved in the data transfer. The maximum segment size being used by TCP for this host is: 1460. Low average size of packets received by this system might be because of a misconfigured client system, or a poorly written client application.
Is someone else seeing such output from netdiag?
I'm trying to locate a performance problem only appearing on linux hosts when the filer is under (heavy) load and a snapmirror proc from a second toaster is started. All linux clients produce a load up to 82 and start to freeze. Of course thats really not nice to have - especially when other bsd/sunos hosts have no problems.
Netapp itself can't help me - they say that the linux nfs drivers are a pieace of crap and that they're working with linux developers to fix that. (blah blah) Someone else ever had such problems? What did you do?
Thanks and greets,
On Tue, 17 Jun 2003 10:30:32 +0100 "Clawson, Simon" simon_clawson@mentorg.com wrote:
What version of Linux? If its 7.3, then do a search on NOW for redhat 7.3+NFS. Its a known bug,a nd the RH Kernel needs upgrading to fix it properly.
All linux hosts run with linux kernel 2.4.18 - one with 2.4.19 (self compiled kernel). The distribution itself is a SuSE8.0.
-----Original Message----- From: Stefan Funke [mailto:bundy@arcor-ip.de] Sent: 17 June 2003 10:15 To: toasters@mathworks.com Subject: netdiag / linux nfs over tcp
Hi all!
Is someone having a lot of linux clients mounted a toaster via nfs over tcp? I have a mixed environment of linux, bsd and sunos clients. When I run "netdiag" here the toaster finds all linux hosts not to perform like they should:
Average size of NFS TCP packets received from host: foo is 461. This is less than the MTU (1500 bytes) of the interface involved in the data transfer. The maximum segment size being used by TCP for this host is: 1460. Low average size of packets received by this system might be because of a misconfigured client system, or a poorly written client application.
Is someone else seeing such output from netdiag?
I'm trying to locate a performance problem only appearing on linux hosts when the filer is under (heavy) load and a snapmirror proc from a second toaster is started. All linux clients produce a load up to 82 and start to freeze. Of course thats really not nice to have - especially when other bsd/sunos hosts have no problems.
Netapp itself can't help me - they say that the linux nfs drivers are a pieace of crap and that they're working with linux developers to fix that. (blah blah) Someone else ever had such problems? What did you do?
i'm not sure but you should have the command mii-tool - type it as root, what do you get ? (for ref here is what I have here : genghis:~# mii-tool eth0: negotiated 100baseTx-FD flow-control, link ok) - what about ifconfig output (type ifconfig <yourdev>) - and finally what about mounts (type mount)
Stefan Funke wrote:
On Tue, 17 Jun 2003 10:30:32 +0100 "Clawson, Simon" simon_clawson@mentorg.com wrote:
What version of Linux? If its 7.3, then do a search on NOW for redhat 7.3+NFS. Its a known bug,a nd the RH Kernel needs upgrading to fix it properly.
All linux hosts run with linux kernel 2.4.18 - one with 2.4.19 (self compiled kernel). The distribution itself is a SuSE8.0.
-----Original Message----- From: Stefan Funke [mailto:bundy@arcor-ip.de] Sent: 17 June 2003 10:15 To: toasters@mathworks.com Subject: netdiag / linux nfs over tcp
Hi all!
Is someone having a lot of linux clients mounted a toaster via nfs over tcp? I have a mixed environment of linux, bsd and sunos clients. When I run "netdiag" here the toaster finds all linux hosts not to perform like they should:
Average size of NFS TCP packets received from host: foo is 461. This is less than the MTU (1500 bytes) of the interface involved in the data transfer. The maximum segment size being used by TCP for this host is: 1460. Low average size of packets received by this system might be because of a misconfigured client system, or a poorly written client application.
Is someone else seeing such output from netdiag?
I'm trying to locate a performance problem only appearing on linux hosts when the filer is under (heavy) load and a snapmirror proc from a second toaster is started. All linux clients produce a load up to 82 and start to freeze. Of course thats really not nice to have - especially when other bsd/sunos hosts have no problems.
Netapp itself can't help me - they say that the linux nfs drivers are a pieace of crap and that they're working with linux developers to fix that. (blah blah) Someone else ever had such problems? What did you do?
On Tue, 17 Jun 2003 12:18:45 +0200 Stephane Bentebba stephane.bentebba@fps.fr wrote:
i'm not sure but you should have the command mii-tool
- type it as root, what do you get ?
(for ref here is what I have here : genghis:~# mii-tool eth0: negotiated 100baseTx-FD flow-control, link ok)
All 100tx-FD, link OK (same at switch side)
- what about ifconfig output (type ifconfig <yourdev>)
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:43303086 errors:0 dropped:0 overruns:0 frame:0 TX packets:48909680 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:100 RX bytes:3674417272 (3504.1 Mb) TX bytes:230870640 (220.1Mb) Interrupt:22 Base address:0xd000
Never seen any errors on the interfaces.
- and finally what about mounts (type mount)
We tried multiple mount options together with netapp (a while ago) - from nfs v2, v3, hard, soft, tcp or udp etc. Right now all hosts have it via nfsv2, tcp, rsize=8192,wsize=8192,intr,bg.
Greetings,
um, . you have a strange problem
i can see 2 way to explore , first the kernel : just for ref, my .config file (extract : grep IP) << genghis:/usr/src/kernel-source-2.4.16# grep IP .config | grep # -v CONFIG_SYSVIPC=y CONFIG_IP_MULTICAST=y CONFIG_NET_IPIP=m CONFIG_NET_IPGRE=m CONFIG_IPX=m CONFIG_SCSI_IPS=y CONFIG_IPDDP=m CONFIG_PLIP=m CONFIG_SLIP=m CONFIG_SLIP_COMPRESSED=y CONFIG_SLIP_SMART=y CONFIG_STRIP=m CONFIG_IPHASE5526=m CONFIG_SERIAL_MULTIPORT=y
second : it could be just normal that you get an mtu to 1460 when packets are the last of a transmission, the last packet can often be trunckated at its usefull size ie be less than 1500
as you talk about 1460 mtu , were you able to make a dump of the nic traffic with dump or ethereal ?
subsidiaries : tcp is good when you have a bad net (i guess you have a bad one as you are in 8192 win size) but why are you still in v2 (surelly some old linux workstation) ?
I have to go out of the office so I wont be able to reply to you today from now .
bye
Stefan Funke wrote:
On Tue, 17 Jun 2003 12:18:45 +0200 Stephane Bentebba stephane.bentebba@fps.fr wrote:
i'm not sure but you should have the command mii-tool
- type it as root, what do you get ?
(for ref here is what I have here : genghis:~# mii-tool eth0: negotiated 100baseTx-FD flow-control, link ok)
All 100tx-FD, link OK (same at switch side)
- what about ifconfig output (type ifconfig <yourdev>)
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:43303086 errors:0 dropped:0 overruns:0 frame:0 TX packets:48909680 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:100 RX bytes:3674417272 (3504.1 Mb) TX bytes:230870640 (220.1Mb) Interrupt:22 Base address:0xd000
Never seen any errors on the interfaces.
- and finally what about mounts (type mount)
We tried multiple mount options together with netapp (a while ago) - from nfs v2, v3, hard, soft, tcp or udp etc. Right now all hosts have it via nfsv2, tcp, rsize=8192,wsize=8192,intr,bg.
Greetings,
-
Clawson, Simon -
Stefan Funke -
Stephane Bentebba