Skip to contentSkip to conversation

nfsroot dead slow with redhat 7.2

DISCUSSION David Chow 3 participants 8 messages Dec 20, 2001

lore
Share & tools
lore

David Chow

author
#0 lore raw

Dear all,

When I use 2.4.7-10 i686 kernel from stock Redhat 7.2 as the NFS server. My NFS client use the 2.4.13 kernel, when I mount the nfsroot to the server, I found it is dead slow on the client. This only happens in i686 kernel on the server, if we use a K6-2 uses an i386 server its fine. What's going on? By the way, how to configure the client to default use a NFSv3 mount? Thanks.

regards,

David

#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == David Chow <[email protected]> writes:

> Dear all, When I use 2.4.7-10 i686 kernel from stock Redhat 7.2 > as the NFS server. My NFS client use the 2.4.13 kernel, when I > mount the nfsroot to the server, I found it is dead slow on the > client. This only happens in i686 kernel on the server, if we > use a K6-2 uses an i386 server its fine. What's going on? By

Usually means you have a bad network connection. Use tcpdump to isolate where on the network packets (and UDP fragments) are disappearing.

> the way, how to configure the client to default use a NFSv3 > mount? Thanks.

Specify the 'v3' NFSroot mount option.

Cheers,
Trond

author
#2 lore raw

Dear Trond,

Thanks for answering my question, we have use the i386 kernel at the server that works fine. Also even we uses the i686 kerenl at the server, it happens normally when doing NFS mounts, it will only be dead slow when

	server i686 2.4.7-10 kernel && client nfsroot(2.4.13 i686)

The network is fine. It is so slow that an ls -l at the rootfs takes more than 2 minutes. The readdir() seems alright because the ls immediate counts the number of records says "total blahbalh" but when doing individual lookup calls, it seems slow like hell. We have other production i686smp servers 2.4.14 serving diskless i686 clients using 2.4.13 kernels works great. Is there any difference in nfsroot with normal nfsmounts? And can we configure the nfsroot use a v3 mount? becaus now it defaults to v2 always.

David

在 週四, 2001-12-20 08:52, Trond Myklebust 寫道:

Quoted 19 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == David Chow <[email protected]> writes:

> Dear all, When I use 2.4.7-10 i686 kernel from stock Redhat 7.2 > as the NFS server. My NFS client use the 2.4.13 kernel, when I > mount the nfsroot to the server, I found it is dead slow on the > client. This only happens in i686 kernel on the server, if we > use a K6-2 uses an i386 server its fine. What's going on? By

Usually means you have a bad network connection. Use tcpdump to isolate where on the network packets (and UDP fragments) are disappearing.

> the way, how to configure the client to default use a NFSv3 > mount? Thanks.

Specify the 'v3' NFSroot mount option.

Cheers,
Trond

#3 lore raw
Quoted 11 lines · David Chow

Dear Trond,

Thanks for answering my question, we have use the i386 kernel at the server that works fine. Also even we uses the i686 kerenl at the server, it happens normally when doing NFS mounts, it will only be dead slow when

	server i686 2.4.7-10 kernel && client nfsroot(2.4.13 i686)

The network is fine. It is so slow that an ls -l at the rootfs takes more than 2 minutes. The readdir() seems alright because the ls immediate counts the number of records says "total blahbalh" but when

Is a "ls -ln" any faster?

Dax

#4 lore raw
In reply to David Chow
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == David Chow <[email protected]> writes:

> The network is fine. It is so slow that an ls -l at the rootfs > takes more than 2 minutes. The readdir() seems alright because > the ls immediate counts the number of records says "total > blahbalh" but when doing individual lookup calls, it seems slow > like hell. We have other production i686smp servers 2.4.14

As I said: 'tcpdump' ought to show you what is going on.

> serving diskless i686 clients using
> 2.4.13 kernels works great. Is there any difference in nfsroot
> with
> normal nfsmounts? And can we configure the nfsroot use a v3

Nope: no differences. NFSroot uses the exact same code as standard NFS.

> mount? becaus now it defaults to v2 always.

As I said in my previous mail: use the mount option 'v3' on the kernel boot line if you want NFSv3.

e.g. nfsroot="10.0.0.1:/bar,v3"

Cheers,
Trond

author
#5 lore raw

在 週四, 2001-12-20 10:39, Trond Myklebust 寫道:

Quoted 20 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == David Chow <[email protected]> writes:

> The network is fine. It is so slow that an ls -l at the rootfs > takes more than 2 minutes. The readdir() seems alright because > the ls immediate counts the number of records says "total > blahbalh" but when doing individual lookup calls, it seems slow > like hell. We have other production i686smp servers 2.4.14

As I said: 'tcpdump' ought to show you what is going on

> serving diskless i686 clients using
> 2.4.13 kernels works great. Is there any difference in nfsroot
> with
> normal nfsmounts? And can we configure the nfsroot use a v3

Nope: no differences. NFSroot uses the exact same code as standard NFS.

> mount? becaus now it defaults to v2 always.

Quoted 7 lines

As I said in my previous mail: use the mount option 'v3' on the kernel boot line if you want NFSv3.

e.g. nfsroot="10.0.0.1:/bar,v3"

Cheers,
Trond

Just find out... it is a problem of some settings in /etc directory it is not related to the FSes . I replaced the /etc directory with the one we are using on the production machines... by the way. What can be wrong? When it starts init , and execute the /etc/rc.d/rc.sysinit , it is hell slow. We have tried replace /sbin/init with bash and we got out a shell but "ls -l" takes more than 2 minutes... do you know what sort of settings in the /etc will affect use space "bash" or "glibc" on nfsroot behaves different ? This is so strange.

Thanks

David

#7 lore raw
In reply to David Chow
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

" " == David Chow <[email protected]> writes:

> Just find out... it is a problem of some settings in /etc > directory it is not related to the FSes . I replaced the /etc > directory with the one we are using on the production > machines... by the way. What can be wrong? When it starts init > , and execute the /etc/rc.d/rc.sysinit , it is hell slow. We > have tried replace /sbin/init with bash and we got out a shell > but "ls -l" takes more than 2 minutes... do you know what sort > of settings in the /etc will affect use space "bash" or "glibc" > on nfsroot behaves different ? This is so strange.

Anything network related: ipchains, network parameters, ... Also, if things like portmapper, statd, and the remaining NFS mounts are getting started in the wrong order.

Cheers,
Trond

#6 lore raw
In reply to David Chow
Quoted 8 lines · David Chow

Just find out... it is a problem of some settings in /etc directory it is not related to the FSes . I replaced the /etc directory with the one we are using on the production machines... by the way. What can be wrong? When it starts init , and execute the /etc/rc.d/rc.sysinit , it is hell slow. We have tried replace /sbin/init with bash and we got out a shell but "ls -l" takes more than 2 minutes... do you know what sort of settings in the /etc will affect use space "bash" or "glibc" on nfsroot behaves different ? This is so strange.

David. I asked yesterday for you to try "ls -ln" and report if it is any faster. What you are just now discovering, I suspected when I first read your report.

Please try "ls -ln" and report.

Dax