Red HatThis forum is for the discussion of Red Hat Linux.
Welcome to LinuxQuestions.org, a friendly and active Linux Community.
You are currently viewing LQ as a guest. By joining our community you will have the ability to post topics, receive our newsletter, use the advanced search, subscribe to threads and access many other special features. Registration is quick, simple and absolutely free. Join our community today!
Note that registered members see fewer ads, and ContentLink is completely disabled once you log in.
Replace the 100Mb/s switch with new 1Gb/s Cisco product, I mount 10T share smoothly,however it still takes about 1 minute to complete "df" operation.
Hence,here comes a new question. Some customers are using 100Mb/s switches and have no plan to buy 1Gb/s switches in recent days, What should I do to help them to execute "df" in their 100Mb/s networking? Or just suggest them to update to faster devices?
nfs using udp instead of tcp
Different MTU sizes set in /etc/sysconfig/network-scripts/ifcfg-eth0
network driver, e.g. IBM blades with tg3 driver instead of broadcomm
Does it work locally, but not remotely?
How are they mounting the directory?
If they don't need to run df on the nfs exported directory, they can use the "-l" switch to show local only. e.g. df -lh
Yes, df works well locally,either 100Mb/s or 1Gb/s networking environment.
The architecture is: four CentOS4.4 act as cluster node, and mount the underlying storage device with gfs option. The outside clients access the NFS share exported in nodes(mount node1:/share client_directory),and all the mount operating use the default options.
The 'mount' time seems to be reasonable, about 5 seconds, but when issue 'df' in 100Mb/s networking, it will hang.
Last edited by PhillipHuang; 05-17-2007 at 03:07 AM.
Perhaps you need to mess around with wsize and rsize values when you run the mount command. Also, mounting it over tcp doesn't hurt. You don't see collisions on the interfaces of the linux servers do you? Are they also at 100Mb full duplex? Do you see any packet errors on the ports of the switches that they are connected to?
I was working in customer fieldground in last friday, and found perhaps underlying reason is related to client's 'mount' version. The NFS server(cluster node) is mount-2.12p, while the client is fairly old, mount-2.11y. When I test in another linux box integrated mount-2.12p, the mount process is very fast, and 'df' does not hang,however, it still takes about 10 seconds to display large shares.
The customer would never upgrade the existing mount in their own Linux system, because it is on their risk. I have to consider to make the node's version lower to adapte the customer.
I will do more test to verify my opinion, and update here.
Last edited by PhillipHuang; 05-19-2007 at 09:48 PM.