does NFS overrides TCP keepalive time?

Question

it seems that TCP keepalive, connection timeout is doing something but the default keepalive time, which I found in /proc/sys/net/ipv4/ is 7200

so i thought tcp keepalive time is overridden by NFS

current problem is that when NFS client takes few minutes to detect server and showing message below, we'd like to shorten this time(No3 from above).

"nfs: server not responding, still trying"

we measured the polling time,

24 -> 48 -> 96 -> 192 -> 300 -> 300

does anyone know where you can find this '300'? or is this because of some other issue rather than TCP keepalive time?

* we already touched kernel parameters from /proc/sys/net/ipv4/tcp_retries2 — aji, Commented Oct 19, 2015 at 4:37
Doesn't the NFS protocol itself detect the absence of the server, instead of relying on the underlying TCP keepalive? — wurtel, Commented Oct 19, 2015 at 9:21
No, NFS is also a protocol, which can use TCP or UDP as transport layer. TCP is an OSI level 4 protocol, NFS is a higher level protocol (one could argue that it's level 5 but there are different opinions about this). TCP in turn uses IP as its underlying protocol. So saying "the protocol is TCP" is al bit short-sighted. So it's possible that NFS is doing some sort of keepalive handshake (i.e. sending TCP data packets to and fro with specific content) which has nothing to do with TCP keepalive. — wurtel, Commented Oct 20, 2015 at 11:15
so is there something that configures TCP keepalive on NFS? sorry for the basical question. thanks — aji, Commented Oct 22, 2015 at 8:24

sendmoreinfo · Accepted Answer · 2016-05-25 23:26:04Z

3

Yes, NFS client does override keepalive settings -- it derives them from 'timeo' mount option. see http://lxr.free-electrons.com/source/net/sunrpc/xprtsock.c for low-level detail.

answered May 25, 2016 at 23:26

sendmoreinfo

2,64322 silver badges37 bronze badges

Add a comment |

1 Answer 1