Bull GNU/Linux NFSv4 project

IPv6 support patches for NFSv4 server for 2.6.27-rc3 kernel . We also supply an IPv6 client and server nfs-utils 1.1.2 package. ... [ Read more ]

Here are the results of a robustness and performances regression tests campaign including benchs runs with kerberos flavors (krb5, krb5i, krb5p) on Intel X86_64 machines. ... [ Read more ]

Here is the Version3 of three commands and a tutorial to facilitate and controlling the setup of a linux kerberized NFS environmment. ... [ Read more ]

Here are the results of a NFSV4 Robustness Tests Campaign running RHEL5.1 Public Gold in front of 2.6.24-rc2-CITI_NFS4_ALL-1 including kerberos on X86_64. ... [ Read more ]

Here is the Version 2 of three commands to facilitate and controlling the setup of a linux kerberized NFS environmment by configuring: - a Kerberos KDC and a Kerberos administration Server - a Kerberos NFS Server - a Kerberos NFS Client ... [ Read more ]

Here is the Version 2 of the tutorial which describes all what need NFSV4 administrators and end-users to easily configure NFSV4 i n a Kerberos environment ... [ Read more ]

Here are the results of a NFSV4 Robustness campaign running SLES10 SP1 in front of 2.6.22-rc5-CITI_NFSV4_ALL-1 on X86_64 64 bits ... [ Read more ]

Here are the results of a robustness and performances regression tests campaign including benchs runs with kerberos flavors (krb5, krb5i, krb5p) on Intel X86_64 machines. ... [ Read more ]

Here is a tutorial which describes all what need NFSV4 administrators and end-users to easily configure NFSV4 in a Kerberos environment. ... [ Read more ]

Here are three commands to facilitate and controlling the setup of a linux kerberized NFS environmment by configuring: - a Kerberos KDC and a Kerberos administration Server - a Kerberos NFS Server - a Kerberos NFS Client ... [ Read more ]

Since NFS does make use of RPC and TI-RPC (for IPv6 now), this test suite is aimed to guarantee that the underlying RPC/TI-RPC technology is robust and reliable. It should also help improving the quality of Linux. ... [ Read more ]

Here are the results of a NFSV4 Robustness campaign running RHEL5 in front of 2.6.19-rc6-CITI_NFSV4_ALL-1 (kerberos included) . ... [ Read more ]

Here are the results of a NFSV4 Robustness campaign running SLES10 in front of 2.6.19-rc6-CITI_NFSV4_ALL-1 (kerberos included) ... [ Read more ]

Here are the results of a robustness and performances regression tests campaign including benchs runs with kerberos flavors (krb5, krb5i, krb5p) on Intel i386 and X86_64 machines. ... [ Read more ]

Several news about TI-RPC:

• the 0.1.7 release has been included in Fedora Core 6 since October. A RPM is now available. Thanks to Steve Diskson from Red Hat
• a version management tool and a bugtracker are now available on new TI-RPC sourceforge (http://sourceforge.net/projects/libtirpc). It will make easier for users to report bug and for developers to maintain versions of the library.

Here are two documents about the migration and replication features in NFSv4. One describes how to configure these features and the other explains a list of tests that have been performed. ... [ Read more ]

Description and results of the security audit of NFSv4. ... [ Read more ]

IPv6 support patches for NFS server are available for 2.6.17 kernel. We also supply an IPv6 ready nfs-utils package (userland server part) ... [ Read more ]

Here is a status of 2.6.17-rc2-CITI_NFS4_ALL-1 with performances results and more resgression tests kerberos included ... [ Read more ]

Here are the first regression tests results on 2.6.17-rc2-CITI_NFS4_ALL-1. It shows this version can be now used rather than previous ones. ... [ Read more ]

Here are the first no-regression tests results on 2.6.16-CITI_NFS4_ALL-2.It shows this version can be now used rather than previous ones. ... [ Read more ]

Here are papers about administration of NFSv4 on Linux. Cover general administration, specific tools and our plans for future development. ... [ Read more ]

Discussions about security issues have shown that the priority was to do tests to avoid memory corruptions problems such as buffer overflows. This page summarize the work done to detect or avoid such problems ... [ Read more ]

NFSv4 core functions are running well under linux-2.6.15 kernel + CITI_NFS4_ALL-1 patch. Morever, the tests show no interoperabity issue between the tested architectures. ... [ Read more ]

The locktest has been parallelized to be run over multiple clients accessing the same server. The test has been successfully with two clients. ... [ Read more ]

This guide aims at helping to migrate from NFSv3 to NFSv4, by explaining for each functions what the limits are ... [ Read more ]

The multiple tests done on NFSv4 with 2.6.15-rc3 kernel shows that there is no regression. This release seems to be stable ... [ Read more ]

Quand il n'y en a plus, il y en a encore... This are the third and fourth lock tests. they have been done with the locktests again. There are few diferences between this and previous tests. The locktest is relauched many times in this configuration. Within this configuration, two bugs have been showed. ... [ Read more ]

This is a second lock test. It as been done with the locktests as in the previous test. However, the kernel is now linux 2.6.14 and Tronds patches have been integrated to fix previous bugs. Short conclusion: Locking works fine ... [ Read more ]

To test and stress NFSv4 locks, a new tool has been written. The locktester program creates a master program and slave programs. The master program opens a file and set a lock on the whole file (fcntl). Then it tells the slaves processes to perform the same operation on the test file. The slave processes try to perform one operation :read the file, write the file, set a write lock, set a read lock... The operation can work or fail. This information is returned to the master program. It compare the result with the expected and predeterminated results of the test Slaves can be created as new processes (fork, default) or as threads (pthread_create, -T). Expected results are differents for some tests ... [ Read more ]

These are a Non-Regression tests done on kernel 2.6.14-rc1. The tests consist in runnning fsx and fsstress. There are several problems on the core network functions. These problems do not look nfsv4-specific. However, it is not possible to complete standard Non-Regression tests without crashing the kernel. ... [ Read more ]

In a previous article, Tony Reix have explained how to setup the netem tool to emulate various network behaviours. In this test, we have setup a configuration of netem to emulate the internet behaviour on a Gigabit local aera network. To validate this configuration we have tried to reproduce a known NFSv4 WAN bug on a the local network. This bug is WAN-specific and is not reproductible on a fast local network, such as our gigabit ethernet network. ... [ Read more ]

We have been able to put the NetEm Network Emulator tool at work with NFSv4 for emulating tests over the WAN. It seems to be a great tool ! ... [ Read more ]

Performance and functionnal tests have been runned between a non-Linux NFS client and a Linux NFS server. ... [ Read more ]

Interoperability tests have been runned between a Linux NFSv4 client and an non-Linux NFS server. These tests include fonctionnal tests (Connectathon) and performance comparison (iozone). ... [ Read more ]

This test tries run NFSv4 over different plateforms. NFSv4 is not ready for 64 bits plateforms. ... [ Read more ]

A bug occurs when trying to set an ACL with more than 35 entries. ... [ Read more ]

A kernel general protection fault occurs when executing the Connectathon locking tests on a file system expoted from a X86 64bits NFS server. ... [ Read more ]

This is a non regression test about interoperablity. The test used is conectathon 2004. Hardware configurations are: * 2x PPC64 machines, 1Go ram * 2x IA32 machines, 2Go ram Lock tests failed with PPC64 Client. NFSv4 client looks to have a bug on PPC64. ... [ Read more ]

The goal of this first test is to know if NFSv4 is robust to share data on Internet. Tests stress the main filesystem functions. Tests used are fsstress and fsx. ... [ Read more ]

The previous exportfs tests shows a performance issue. A patch has been written. The patch changes the the simple chained list of exported directories into a hashed table. This test tries to measure the influence of the patch on exportfs performance. ... [ Read more ]

This test try to show the maximal number of connections a single NFSv4/ia32/Linux server is able to accept. The server export one directory (server:/). The difference between this test and the previous one is about the load of the server. In fact, if the two tests mount an unique exported directory on multiple mount points, this test launch an IOZone benchmark on each nfs4 share. ... [ Read more ]

After having analyzed the performance of NFSv4 for files creation, compared it to ext3 performance, and compared it to NFSv3 performance, we have concluded that there is a performance issue in NFSv4. To help to find where the problem is, open is done with O_CREATE | O_RDWR and not O_CREATE alone. ... [ Read more ]

This test try to show the maximal number of connections a single NFSv4/ia32/Linux server is able to accept is able to accept. The server export one directory (server:/). The client create a mount point and mount the server export on to the mount point. It repeat the operation as many time as possible until it fails. In this test, client and servers are on the same machine, using the loopback. Time to perform each mount is measured and reported. Limit is at 31998 mounts. ... [ Read more ]

After having analyzed the performance of NFSv4 for files creation, compared it to ext3 performance, and compared it to NFSv3 performance, we have concluded that there is a performance issue in NFSv4. To help to find where the problem is, I have compared the performance of NFSv4 on Linux with a non-Linux server. ... [ Read more ]

A new IPv6 support patchset is available for 2.6.12 kernel. It includes change related to new kernel release. The parts of code of the previous release related to rpcbind client have been integrated into Chuck Lever's transport switch. ... [ Read more ]

After having analysed performances of NFSv4 for files creation, and compared it to ext3 performances, the next step is to compare NFSv4 to NFSv3 for files creation. ... [ Read more ]

Chuck Lever's Transport Switch patchset for 2.6.12 kernel. These patches provide in-kernel rpcbind version 3 and 4 client, including rpcbind related modifications from Bull IPv6 support patchset ... [ Read more ]

NFSv4 file creation compared to ext3 file creation. Conclusion : Local filesystem is not responsable for NFSv4 limitations. ... [ Read more ]

To find the maximal number of file NFSv4 is able to manage, I create as many files as possible, and I measure the time of creation. When a directory reach 1,620,000 I note a clear change of behaviour. ... [ Read more ]

After some bugs fixes from last test, NFSv4 is able to list more than 200,000 files. Time to list a directory is O(n2) ... [ Read more ]

How many files NFSv4 is able to manage? The first point is to know 'how many files NFSv4 is able to stat in a directory'... ... [ Read more ]

More details about exportfs, time to export one directory and limits. ... [ Read more ]

Modification after reviewers comments, big patche splited into smaller one. New test plan done a results avalaibles- 'mini how-to' to enable IPv6 support ... [ Read more ]

A list of last bugs found around NFSv4 and kerberos last weeks. ... [ Read more ]

I tried to found the maximum number of file system exported and mounted a single client and server can do ... [ Read more ]

The maximal file size over NFSv4 is the maximal file size the local file system of the server is able to manage. For XFS on ia32 - that is 8TB. ... [ Read more ]

FFSB tool show a robustness lack on the knfsd. Patches are avalaibles. ... [ Read more ]

Regression : A long sequence of mount - umount causes kernel bug. ... [ Read more ]

To help developments client's root does not a valid kerberos ticket to access the server. ... [ Read more ]

IPv6 support implementation is available for the client part. You can now dowload separate or cumulative patches ... [ Read more ]

NFSv4 has been run under several stress tools with one client during more than 24 hours without any problem! NFSv4 core functions can be considered as stable. ... [ Read more ]

A new version for the IPv6 support including remarks and changes in server implementation ... ... [ Read more ]

The first version of the NFSv4 Linux scalability test plan. ... [ Read more ]

This document is for administrator who wants to administrate NFSv4 with Webmin and Nagios. ... [ Read more ]

CITI_NFS4_ALL-5 patches are tested. There are few changes from CITI_NFS4_ALL-3 patches. ... [ Read more ]

New patchset implementing the IPv6 support in the NFS client is available ... [ Read more ]

A long list of patches to correct most of bugs found with the last kernel. However, some are remaining ... [ Read more ]

A complete implementation of NFS over IPv6 has been done and successfully tested. However,this implementation does not meet a general agreement. ... [ Read more ]

fsstress runs multiple process performing random sequences of operations over NFSv4 and showing multiples bugs. ... [ Read more ]

The Nagios plugin 'check_nfs4.pl' allows you to monitor the main features of NFSv4. ... [ Read more ]

Last time I have compared standard NFSv3 to the standard NFSv4. More tests show interresting results, when changing the transport protocol... ... [ Read more ]

Testing NFS with 2.6.10 - 2.6.11 kernel show interoperability problems, dues to an ethernet driver bug. ... [ Read more ]

NFSv4 is tested over ext3, reiser3,xfs, jfs. There is no significative performances diference between each local filesystem ... [ Read more ]

NFSv3 performances are better, principally because we had used NFSv3 over UDP. NFSv4 will be tested versus NFSv3 over TCP. ... [ Read more ]

No performances regression on kernel 2.6.10 ... [ Read more ]