Results 1 to 2 of 2
Enjoy an ad free experience by logging in. Not a member yet? Register.
- Join Date
- Apr 2006
Poor NFS Write Performance, System Hangs
I tested the speed using:
time dd if=/dev/zero of=/mnt/ddtest bs=16k count=1638400
time dd if=/mnt/ddtest of=/dev/null bs=16k
I tried various settings for rsize/wsize (131072, 65563, 32767, 8192) with little change (actually degraded slightly with each smaller size). The default 128K (131072) seems to be the best. Larger amounts defaulted back to 128K so I assume that is the maximum supported value.
I performed the same test locally on the client and server machines and achieved hard drive read/write rates in the 70MB/sec range so the bottle neck seems to be the network.
With scp I get 14MB/sec on both read and write. I'm not surprised the scp read rate is lower than NFS due to potential ssh overhead, but was happy to see that the read and write rates were the same as I thought they should be since my network is full duplex.
In addition to the slow write rate, my server seems to hang intermittently under NFS stress (dd read of 25G files ) requiring reboot to recover. Unsure where to look for messages for any hints as to why.
With scp I get 14MB/sec on both read and write.
- Join Date
- Nov 2007
happy to see that the read and write rates were the same as I thought they should be since my network is full duplex.
You'll need to do more research to understand the ways that a system will buffer network transfers. If I send 100MB to another machine, it will easily fit in memory and the receiving machine says "data received." But when the buffer runs out and the system has to start sending data to HDD, the speed can be affected greatly.
The fast the system hangs also points to a driver/HW problem.