Matrix robustness section
From Linux NFS
(Difference between revisions)
(→Scalability (robustness)) |
(→Recovery from problems while under light/normal/heavy loads) |
||
(37 intermediate revisions not shown) | |||
Line 4: | Line 4: | ||
==Basic stability assessments== | ==Basic stability assessments== | ||
- | {|border="1" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | + | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% |
!style="background: #ececec;"|'''ID | !style="background: #ececec;"|'''ID | ||
!style="background: #ececec;"|'''test''' | !style="background: #ececec;"|'''test''' | ||
Line 26: | Line 26: | ||
*Large number of maps | *Large number of maps | ||
*Randomly mount and run workloads on an automounted partition | *Randomly mount and run workloads on an automounted partition | ||
- | *use a variety of workloads, such as randomly chosen fs tests | + | *use a variety of workloads, such as randomly chosen fs tests |
|e.g. Crashme [http://people.delphiforums.com/gjc/crashme.html more] | |e.g. Crashme [http://people.delphiforums.com/gjc/crashme.html more] | ||
|'''New''' | |'''New''' | ||
Line 34: | Line 34: | ||
|III.A.3 | |III.A.3 | ||
|Run NFS server for 2 wks with random configuration changes, using: | |Run NFS server for 2 wks with random configuration changes, using: | ||
- | *Interrupt server in various ways (reboot, power cycle, lan fail) | + | *Interrupt server in various ways (reboot, power cycle, lan fail) |
- | *Change/reexport export rules at random | + | *Change/reexport export rules at random |
- | *Trigger a client workload at arbitrary times | + | *Trigger a client workload at arbitrary times |
*Analyze client recovery behaviors | *Analyze client recovery behaviors | ||
| | | | ||
Line 45: | Line 45: | ||
|III.A.4 | |III.A.4 | ||
|Run connectathon locking tests against NFS server for 2 weeks, using: | |Run connectathon locking tests against NFS server for 2 weeks, using: | ||
- | *Multiple client machines | + | *Multiple client machines |
*Reboot at random | *Reboot at random | ||
- | *Analyze client cache coherency behaviors | + | *Analyze client cache coherency behaviors |
*Analyze locking behaviors | *Analyze locking behaviors | ||
| | | | ||
Line 72: | Line 72: | ||
|1 day | |1 day | ||
|} | |} | ||
+ | |||
+ | |||
== Resource limit testing == | == Resource limit testing == | ||
- | {|border="1" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | + | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% |
!style="background: #ececec;"|'''ID | !style="background: #ececec;"|'''ID | ||
!style="background: #ececec;"|'''test''' | !style="background: #ececec;"|'''test''' | ||
Line 155: | Line 157: | ||
==Stress load testing== | ==Stress load testing== | ||
- | {|border="1" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | + | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% |
!style="background: #ececec;"|'''ID | !style="background: #ececec;"|'''ID | ||
!style="background: #ececec;"|'''test''' | !style="background: #ececec;"|'''test''' | ||
Line 186: | Line 188: | ||
==Scalability (robustness)== | ==Scalability (robustness)== | ||
- | {|border="1" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | + | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% |
!style="background: #ececec;"|'''ID | !style="background: #ececec;"|'''ID | ||
!style="background: #ececec;"|'''test''' | !style="background: #ececec;"|'''test''' | ||
Line 195: | Line 197: | ||
|- | |- | ||
|III.D.1 | |III.D.1 | ||
+ | Part 1 : without load | ||
|Find maximum number of connections to Linux IA-32 server | |Find maximum number of connections to Linux IA-32 server | ||
|Fsstress, fsx | |Fsstress, fsx | ||
- | |''' | + | |'''In progress''' |
|Bull (partial) | |Bull (partial) | ||
- | | | + | |Maximum number of connections to a Linux IA-32 server is '''31998'''. ''Warning'' this connections are done by a simple mount on the client. The exported file system is mounted on diferent directories until an error is reported. In this test client and server are on the same machine and the error reported is ''[Errno 31] Too many links'' - See III.D.4 test and [http://nfsv4.bullopensource.org/tools/tests/page25.php test results] |
+ | |- | ||
+ | |III.D.1 | ||
+ | Part 2 : with load | ||
+ | |Find maximum number of connections to Linux IA-32 server | ||
+ | |IOZone | ||
+ | |'''In progress''' | ||
+ | |Bull (partial) | ||
+ | |A single bi IA32 processor is able to answer 2048 client with hight load. 2048 simultaneous instances of IOzone's standard tests were launched. They end sucessfully after 6 days. (Note: with only one IOzone instance time to perform the test is about 45minutes, depending on the test conditions). '''[http://nfsv4.bullopensource.org/tools/tests/page26.php Read more]''' | ||
|- | |- | ||
|III.D.2 | |III.D.2 | ||
|Find maximum number of files for Linux IA-32 exported file system | |Find maximum number of files for Linux IA-32 exported file system | ||
- | | | + | |[http://nfsv4.bullopensource.org/tools/tests_tools/test_files.py Addhoc tool] |
- | |''' | + | |'''Done''' |
- | | | + | |Bull |
- | | | + | |A first [http://nfsv4.bullopensource.org/tools/tests/page18.php test] give a limit between 15,699 and 15,799. Some patches after that a [http://nfsv4.bullopensource.org/tools/tests/page19.php new limit] is over 250,000 files. This limit is the maximal number of files NFS is able to list in a shared directory. The time creation was [http://nfsv4.bullopensource.org/tools/tests/page20.php measured] and there is a change of behaviour when number of files is over 1,620,000 in a directory. This limitation is now fixed (linux 2.6.13 +) |
|- | |- | ||
|III.D.3 | |III.D.3 | ||
|Find maximum file size on Linux IA-32 | |Find maximum file size on Linux IA-32 | ||
- | | | + | |[http://nfsv4.bullopensource.org/tools/tests_tools/bigfile.tar.gz adhoc tool] |
- | |''' | + | |'''Done''' |
- | | | + | |Bull |
- | | | + | |size is the maximal size the server's local filesystem is able to manage. 8TB for NFSv4/XFS/ia32. |
|- | |- | ||
|III.D.4 | |III.D.4 | ||
|Find maximum number of mounted file systems on client | |Find maximum number of mounted file systems on client | ||
- | | | + | |[http://nfsv4.bullopensource.org/tools/tests_tools/test_mount.py adhoc tool] |
- | |''' | + | |'''done''' |
|Bull | |Bull | ||
- | | | + | |Maximum number of connections to Linux IA-32 server is '''31998'''. ''Warning'' this connections are done by a simple mount on the client. The exported file system is mounted on diferent directories until an error is reported. In this test client and server are on the same machine and the error reported is ''[Errno 31] Too many links'' see [http://nfsv4.bullopensource.org/tools/tests/page25.php test results] |
|- | |- | ||
|III.D.5 | |III.D.5 | ||
Line 245: | Line 256: | ||
|III.D.8 | |III.D.8 | ||
|Find maximum number exported file systems on server | |Find maximum number exported file systems on server | ||
+ | |[http://nfsv4.bullopensource.org/tools/tests_tools/test_mount.py adhoc tool] | ||
+ | |'''Near done''' | ||
+ | |Bull | ||
+ | |No limit reached, but the export process is too slow to be acceptable after 3000 exports. 13000 exports reached. See [http://nfsv4.bullopensource.org/tools/tests/page17.php this] page for more information. | ||
+ | |- | ||
+ | |III.D.9 | ||
+ | |Find maximum size of exported file systems on server | ||
+ | | | ||
+ | |'''Done''' | ||
+ | | | ||
+ | |Maximum size of exported file system on server do not depend on NFS/NFS tools. NFSv4 tools do not use filesystem size informations. Limit is the file system limit it self. | ||
+ | |- | ||
+ | |III.D.10 | ||
+ | |Access a single locked file with a large amount of clients | ||
+ | |locktester | ||
+ | |'''Done''' | ||
+ | |'''Bull''' | ||
+ | |A lot of bug fixes. Working [http://nfsv4.bullopensource.org/tools/tests/page41.php now]. | ||
+ | |- | ||
+ | |III.D.11 | ||
+ | |Divide a file into a hight number of locked sections | ||
+ | |locktester | ||
+ | |'''Done''' | ||
+ | |'''Bull''' | ||
+ | |A lot of bug fixes. Working [http://nfsv4.bullopensource.org/tools/tests/page41.php now]. | ||
+ | |- | ||
+ | |} | ||
+ | |||
+ | ==Recovery from problems while under light/normal/heavy loads== | ||
+ | |||
+ | |||
+ | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | ||
+ | !style="background: #ececec;"|'''ID | ||
+ | !style="background: #ececec;"|'''test''' | ||
+ | !style="background: #ececec;"|'''tool test''' | ||
+ | !style="background: #ececec;"|'''status''' | ||
+ | !style="background: #ececec;"|'''owner''' | ||
+ | !style="background: #ececec;"|'''notes''' | ||
+ | |- | ||
+ | |III.E.1 | ||
+ | |Test short & long term local network failure (unplugged cable, ifdown eth0, etc.) | ||
+ | | | ||
+ | |'''Open''' | ||
+ | |OSDL | ||
+ | | | ||
+ | |- | ||
+ | |III.E.2 | ||
+ | |Test short & long duration remote network partition | ||
+ | | | ||
+ | |'''Open''' | ||
+ | |OSDL | ||
+ | | | ||
+ | |- | ||
+ | |III.E.3 | ||
+ | |Test behavior during crash/reboot of server with clients holding various states | ||
+ | | | ||
+ | |'''Open''' | ||
+ | |OSDL | ||
+ | |[ftp://ftp.cis.uoguelph.ca/pub/nfsv4/testing-stuff more] | ||
+ | |- | ||
+ | |III.E.4 | ||
+ | |Test multiple clients using, locking, etc. same files | ||
+ | | | ||
+ | |'''Done''' | ||
+ | |Bull | ||
+ | |[http://nfsv4.bullopensource.org/tools/tests/page41.php More] | ||
+ | |- | ||
+ | |III.E.5 | ||
+ | |Test behavior of server with failed storage device | ||
| | | | ||
|'''New''' | |'''New''' | ||
Line 250: | Line 330: | ||
| | | | ||
|- | |- | ||
- | |III. | + | |III.E.6 |
- | | | + | |Test behavior during crash of client with open delegations and locks |
+ | | | ||
+ | |'''New''' | ||
+ | | | ||
+ | | | ||
+ | |- | ||
+ | |III.E.7 | ||
+ | |Test recovery from denied permission | ||
+ | | | ||
+ | |'''New''' | ||
+ | | | ||
+ | | | ||
+ | |- | ||
+ | |III.E.8 | ||
+ | |Test recovery from JUKEBOX/DELAY | ||
+ | | | ||
+ | |'''New''' | ||
+ | | | ||
+ | | | ||
+ | |- | ||
+ | |III.E.9 | ||
+ | |Test recovery from ESTALE | ||
| | | | ||
|'''New''' | |'''New''' | ||
Line 258: | Line 359: | ||
|} | |} | ||
- | == | + | ==Race conditions== |
- | + | ||
- | + | ||
- | + | ||
- | + | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | |
- | + | !style="background: #ececec;"|'''ID | |
- | + | !style="background: #ececec;"|'''test''' | |
- | + | !style="background: #ececec;"|'''tool test''' | |
- | + | !style="background: #ececec;"|'''status''' | |
- | + | !style="background: #ececec;"|'''owner''' | |
- | + | !style="background: #ececec;"|'''notes''' | |
- | + | |- | |
- | + | |III.F.1 | |
- | + | |Test for race conditions and locking bugs on PPC64 | |
- | + | | | |
- | + | |'''New''' | |
- | + | |(Polyserve?) | |
- | Test | + | |Olaf Kirch says PPC64 is good at exposing problems because of its weak CPU cache coherency semantics |
- | + | |- | |
- | + | |III.F. | |
- | + | |Test for race conditions on new architectures | |
- | + | | | |
- | + | |'''New''' | |
- | + | |(Polyserve?) | |
- | + | |Faster CPU, memory, and buses can expose race conditions | |
- | + | |} | |
- | + | ||
- | + | ==Automounter robustness== | |
- | + | ||
- | + | For more info about Automounter, see notes in nfsv4 list archive for 2/16/05 | |
- | + | ||
- | + | {|border="1" width="100%" cellpadding="1" cellspacing="0" style="font-size: 85%; border: gray solid 1px; border-collapse: collapse; text-align: center; width: 100% | |
- | + | !style="background: #ececec;"|'''ID | |
- | + | !style="background: #ececec;"|'''test''' | |
- | + | !style="background: #ececec;"|'''tool test''' | |
- | + | !style="background: #ececec;"|'''status''' | |
- | + | !style="background: #ececec;"|'''owner''' | |
- | + | !style="background: #ececec;"|'''notes''' | |
- | + | |- | |
- | + | |III.G.1 | |
- | + | |Test interuptible automounting in the following cases | |
- | + | *indirect mount | |
- | Test | + | *direct mount |
- | + | *browsed mount | |
- | + | *multimount offset | |
- | + | | | |
- | + | |'''New''' | |
- | + | | | |
- | + | | | |
- | + | |- | |
- | + | |III.G.2 | |
- | + | |Test concurrent access tests for races in automounter | |
- | + | *Have multiple threads working in parallel | |
- | + | | | |
- | + | |'''New''' | |
+ | | | ||
+ | | | ||
+ | |- | ||
+ | |III.G.3 | ||
+ | |Test replicated file system selection | ||
+ | | | ||
+ | |'''New''' | ||
+ | | | ||
+ | | | ||
+ | |- | ||
+ | |III.G.4 | ||
+ | |Test remounting after expire corner cases: | ||
+ | *Something (a process) sitting in the scaffolding | ||
+ | *Common case for /net | ||
+ | | | ||
+ | |'''New''' | ||
+ | | | ||
+ | |Needs to be supported at nfs level | ||
+ | |} |
Latest revision as of 11:37, 9 November 2005
Section III
Contents |
ROBUSTNESS TESTING
Basic stability assessments
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.A.1 | Run iozone for 2 weeks on basic client/server operations, using:
| IOzone | done | BULL | Now testing with fsstress and FFSB |
III.A.2 | Run automounter use case for 2 weeks on amd, autofs, and autong, using:
| e.g. Crashme more | New | ||
III.A.3 | Run NFS server for 2 wks with random configuration changes, using:
| OPEN | OSDL | ||
III.A.4 | Run connectathon locking tests against NFS server for 2 weeks, using:
| NEW | |||
III.A.5 | Run fsstress 2 weeks on basic client/server operations, using:
| fsstress | DONE | BULL | 1 week |
III.A.6 | Run FFSB 1 day on basic client/server operations in stress configuration, using:
| ffsb | DONE | BULL | 1 day |
Resource limit testing
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.B.1 | Test stability of client in out of pid situation | ||||
III.B.2 | Test stability of client in out of memory situation | valgrind | New | ||
III.B.3 | Test stability of client in out of disk space on server situation | dd,fsstress | Done | Bull | Simple error message no space left on device |
III.B.4 | Test stability of client in out of inode situation | ||||
III.B.5 | Test stability of client in out of swap space situation | ||||
III.B.6 | Test stability of server in out of pid situation | ||||
III.B.7 | Test stability of server in out of memory situation | valgrind | New | ||
III.B.8 | Test stability of server in out of disk space | dd,fsstress | Done | Bull | Simple error message no space left on device |
III.B.9 | Test stability of server in out of inode situation | ||||
III.B.10 | Test stability of server in out of swap space situation |
Stress load testing
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.C.1 | Run stress tools in a std config on each release | fsx,fsstress,ffsb | In progress | BULL | fsstress and ffsb are ran 1hour |
III.C.2 | Analyze load balancing, failure modes, etc. under different stress loads | New | |||
III.C.3 | Destructive testing by measuring point of failure for various loads | New |
Scalability (robustness)
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.D.1
Part 1 : without load | Find maximum number of connections to Linux IA-32 server | Fsstress, fsx | In progress | Bull (partial) | Maximum number of connections to a Linux IA-32 server is 31998. Warning this connections are done by a simple mount on the client. The exported file system is mounted on diferent directories until an error is reported. In this test client and server are on the same machine and the error reported is [Errno 31] Too many links - See III.D.4 test and test results |
III.D.1
Part 2 : with load | Find maximum number of connections to Linux IA-32 server | IOZone | In progress | Bull (partial) | A single bi IA32 processor is able to answer 2048 client with hight load. 2048 simultaneous instances of IOzone's standard tests were launched. They end sucessfully after 6 days. (Note: with only one IOzone instance time to perform the test is about 45minutes, depending on the test conditions). Read more |
III.D.2 | Find maximum number of files for Linux IA-32 exported file system | Addhoc tool | Done | Bull | A first test give a limit between 15,699 and 15,799. Some patches after that a new limit is over 250,000 files. This limit is the maximal number of files NFS is able to list in a shared directory. The time creation was measured and there is a change of behaviour when number of files is over 1,620,000 in a directory. This limitation is now fixed (linux 2.6.13 +) |
III.D.3 | Find maximum file size on Linux IA-32 | adhoc tool | Done | Bull | size is the maximal size the server's local filesystem is able to manage. 8TB for NFSv4/XFS/ia32. |
III.D.4 | Find maximum number of mounted file systems on client | adhoc tool | done | Bull | Maximum number of connections to Linux IA-32 server is 31998. Warning this connections are done by a simple mount on the client. The exported file system is mounted on diferent directories until an error is reported. In this test client and server are on the same machine and the error reported is [Errno 31] Too many links see test results |
III.D.5 | Test robustness on NUMA when scaling CPU, mem, NIC, or disk count | Fsstress, fsx | New | ||
III.D.6 | Test robustness on SMP when scaling CPU, mem, NIC, or disk count | Fsstress, fsx | New | Bull (partial) | |
III.D.7 | Test correctness of NFS client when backed by a large (>100GB) cachefs | New | |||
III.D.8 | Find maximum number exported file systems on server | adhoc tool | Near done | Bull | No limit reached, but the export process is too slow to be acceptable after 3000 exports. 13000 exports reached. See this page for more information. |
III.D.9 | Find maximum size of exported file systems on server | Done | Maximum size of exported file system on server do not depend on NFS/NFS tools. NFSv4 tools do not use filesystem size informations. Limit is the file system limit it self. | ||
III.D.10 | Access a single locked file with a large amount of clients | locktester | Done | Bull | A lot of bug fixes. Working now. |
III.D.11 | Divide a file into a hight number of locked sections | locktester | Done | Bull | A lot of bug fixes. Working now. |
Recovery from problems while under light/normal/heavy loads
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.E.1 | Test short & long term local network failure (unplugged cable, ifdown eth0, etc.) | Open | OSDL | ||
III.E.2 | Test short & long duration remote network partition | Open | OSDL | ||
III.E.3 | Test behavior during crash/reboot of server with clients holding various states | Open | OSDL | more | |
III.E.4 | Test multiple clients using, locking, etc. same files | Done | Bull | More | |
III.E.5 | Test behavior of server with failed storage device | New | |||
III.E.6 | Test behavior during crash of client with open delegations and locks | New | |||
III.E.7 | Test recovery from denied permission | New | |||
III.E.8 | Test recovery from JUKEBOX/DELAY | New | |||
III.E.9 | Test recovery from ESTALE | New |
Race conditions
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.F.1 | Test for race conditions and locking bugs on PPC64 | New | (Polyserve?) | Olaf Kirch says PPC64 is good at exposing problems because of its weak CPU cache coherency semantics | |
III.F. | Test for race conditions on new architectures | New | (Polyserve?) | Faster CPU, memory, and buses can expose race conditions |
Automounter robustness
For more info about Automounter, see notes in nfsv4 list archive for 2/16/05
ID | test | tool test | status | owner | notes |
---|---|---|---|---|---|
III.G.1 | Test interuptible automounting in the following cases
| New | |||
III.G.2 | Test concurrent access tests for races in automounter
| New | |||
III.G.3 | Test replicated file system selection | New | |||
III.G.4 | Test remounting after expire corner cases:
| New | Needs to be supported at nfs level |