Skip to contentSkip to conversation

if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

PATCH New Petr Vorel 2 participants 7 messages Jan 7, 2021 – Jan 14, 2021

lore
Share & tools
lore

Petr Vorel

author
#0 lore raw

PATCH 1/1 if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

to make testing faster.

Tested only on netns based testing.

---
Hi Alexey,

any reason why CHANGE_INTERVAL was set 5s?

It'd be nice to speedup the tests, which were slow even before 2d422edbf ("if-mtu-change.sh: Add max packet size detection for IPv4") which added 25% slowdown.

Could you please test this on two host based setup?

Kind regards,
Petr

 testcases/network/stress/interface/if-mtu-change.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/network/stress/interface/if-mtu-change.sh +1 -1
index d2816606b..3efe00461 100755
@@ -11,7 +11,7 @@ TST_CLEANUP="do_cleanup"
 . if-lib.sh
 
 # The interval of the mtu change [second]
-CHANGE_INTERVAL=${CHANGE_INTERVAL:-5}
+CHANGE_INTERVAL=${CHANGE_INTERVAL:-1}
 
 TST_TIMEOUT=$(((CHANGE_INTERVAL + 30) * MTU_CHANGE_TIMES))
 
Signature · 1 line
2.29.2
#1 lore raw

PATCH 1/1 if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

Quoted 4 lines · Petr Vorel

to make testing faster.

Tested only on netns based testing.

Hi Petr,

Quoted 5 lines

---
Hi Alexey,

any reason why CHANGE_INTERVAL was set 5s?

It's more or less safe time for default setup. Not sure why we don't have tst_sleep in if_updown.sh though. I think this is for preventing link-flap errors on the switch...

For netns it's can be set far less of cause by overriding CHANGE_INTERVAL.

Quoted 17 lines

It'd be nice to speedup the tests, which were slow even before 2d422edbf ("if-mtu-change.sh: Add max packet size detection for IPv4") which added 25% slowdown.

Could you please test this on two host based setup?

Kind regards,
Petr

 testcases/network/stress/interface/if-mtu-change.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/network/stress/interface/if-mtu-change.sh +0 -0
index d2816606b..3efe00461 100755

testcases/network/stress/interface/if-mtu-change.sh TST_CLEANUP="do_cleanup" · line 16

Quoted 7 lines
testcases/network/stress/interface/if-mtu-change.sh +1 -1
@@ -11,7 +11,7 @@ TST_CLEANUP="do_cleanup"
 . if-lib.sh
 
 # The interval of the mtu change [second]
-CHANGE_INTERVAL=${CHANGE_INTERVAL:-5}
+CHANGE_INTERVAL=${CHANGE_INTERVAL:-1}
 
 TST_TIMEOUT=$(((CHANGE_INTERVAL + 30) * MTU_CHANGE_TIMES))

It's better to remove TST_TIMEOUT so that CHANGE_INTERVAL can be set, for example, to "100ms" for netns setup.

author
#2 lore raw

PATCH 1/1 if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

Hi Alexey,

Quoted 12 lines · Petr Vorel
Quoted 3 lines

to make testing faster.

Tested only on netns based testing.

Hi Petr,

Quoted 3 lines

---
Hi Alexey,

Quoted 5 lines
Quoted 1 line

any reason why CHANGE_INTERVAL was set 5s?

It's more or less safe time for default setup. Not sure why we don't have tst_sleep in if_updown.sh though. I think this is for preventing link-flap errors on the switch...

Thanks for info.

BTW do you consider sleep $NS_DURATION as needed in testcases/network/stress/multicast/{packet-flood,query-flood}? It's before killing utils (ns-mcast_receiver, ns-udpsender), thus it could be removed during rewrite [1]

Quoted 25 lines

For netns it's can be set far less of cause by overriding CHANGE_INTERVAL.

Quoted 22 lines

It'd be nice to speedup the tests, which were slow even before 2d422edbf ("if-mtu-change.sh: Add max packet size detection for IPv4") which added 25% slowdown.

Could you please test this on two host based setup?

Kind regards,
Petr

 testcases/network/stress/interface/if-mtu-change.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/network/stress/interface/if-mtu-change.sh +1 -1
index d2816606b..3efe00461 100755
@@ -11,7 +11,7 @@ TST_CLEANUP="do_cleanup"
 . if-lib.sh
 
 # The interval of the mtu change [second]
-CHANGE_INTERVAL=${CHANGE_INTERVAL:-5}
+CHANGE_INTERVAL=${CHANGE_INTERVAL:-1}
Quoted 5 lines

testcases/network/stress/interface/if-mtu-change.sh TST_CLEANUP="do_cleanup" · line 16

Quoted 1 line
testcases/network/stress/interface/if-mtu-change.sh +0 -0
 TST_TIMEOUT=$(((CHANGE_INTERVAL + 30) * MTU_CHANGE_TIMES))

It's better to remove TST_TIMEOUT so that CHANGE_INTERVAL can be set, for example, to "100ms" for netns setup.

How about keeping it, but consider CHANGE_INTERVAL as 1 if not a number (i.e. containing "ms", check with tst_is_int would be IMHO enough). I'll send a patch.

Kind regards,
Petr

[1] https://patchwork.ozlabs.org/project/ltp/list/?series=216562

#3 lore raw

PATCH 1/1 if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

Quoted 19 lines · Petr Vorel

Hi Alexey,

Quoted 12 lines · Petr Vorel
Quoted 3 lines

to make testing faster.

Tested only on netns based testing.

Hi Petr,

Quoted 3 lines

---
Hi Alexey,

Quoted 5 lines
Quoted 1 line

any reason why CHANGE_INTERVAL was set 5s?

It's more or less safe time for default setup. Not sure why we don't have tst_sleep in if_updown.sh though. I think this is for preventing link-flap errors on the switch...

Quoted 6 lines

Thanks for info.

BTW do you consider sleep $NS_DURATION as needed in testcases/network/stress/multicast/{packet-flood,query-flood}? It's before killing utils (ns-mcast_receiver, ns-udpsender), thus it could be removed during rewrite [1]

Hi Petr,

Do you mean "sleep $NS_DURATION" in mcast-pktfld02.sh is not needed or something else?

Quoted 40 lines
Quoted 26 lines

For netns it's can be set far less of cause by overriding CHANGE_INTERVAL.

Quoted 22 lines

It'd be nice to speedup the tests, which were slow even before 2d422edbf ("if-mtu-change.sh: Add max packet size detection for IPv4") which added 25% slowdown.

Could you please test this on two host based setup?

Kind regards,
Petr

 testcases/network/stress/interface/if-mtu-change.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/network/stress/interface/if-mtu-change.sh +1 -1
index d2816606b..3efe00461 100755
@@ -11,7 +11,7 @@ TST_CLEANUP="do_cleanup"
 . if-lib.sh
 
 # The interval of the mtu change [second]
-CHANGE_INTERVAL=${CHANGE_INTERVAL:-5}
+CHANGE_INTERVAL=${CHANGE_INTERVAL:-1}
Quoted 5 lines

testcases/network/stress/interface/if-mtu-change.sh TST_CLEANUP="do_cleanup" · line 16

Quoted 1 line
testcases/network/stress/interface/if-mtu-change.sh +0 -0
 TST_TIMEOUT=$(((CHANGE_INTERVAL + 30) * MTU_CHANGE_TIMES))

It's better to remove TST_TIMEOUT so that CHANGE_INTERVAL can be set, for example, to "100ms" for netns setup.

How about keeping it, but consider CHANGE_INTERVAL as 1 if not a number (i.e. containing "ms", check with tst_is_int would be IMHO enough). I'll send a patch.

Kind regards,
Petr

[1] https://urldefense.com/v3/__https://patchwork.ozlabs.org/project/ltp/list/?series=216562__;!!GqivPVa7Brio!J2u5nbibRS-k2rNR25GEw-G5q5x-tv_QzNwnbwDSga6KF3fsu3AeXSBKNL9IjIsvK6aV$

#5 lore raw

PATCH 1/1 if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

Quoted 9 lines · Petr Vorel

Hi Alexey,

Quoted 3 lines

Hi Petr,

Do you mean "sleep $NS_DURATION" in mcast-pktfld02.sh is not needed

Yes: sleep $NS_DURATION in mcast-pktfld02.sh and other mcast*.sh tests in:
https://urldefense.com/v3/__https://patchwork.ozlabs.org/project/ltp/patch/[email protected]/__;!!GqivPVa7Brio!P0xbdqXlRQX7GPsKJmcPapdEOX5VV4OnkCKv-P9OqIWNcTvY9a9KXhOSjdJ1dKco5YbG$
https://urldefense.com/v3/__https://patchwork.ozlabs.org/project/ltp/patch/[email protected]/__;!!GqivPVa7Brio!P0xbdqXlRQX7GPsKJmcPapdEOX5VV4OnkCKv-P9OqIWNcTvY9a9KXhOSjdJ1dKjzIVpO$

It's needed because the commands run in parallel in the background during $NS_DURATION. Without it, the test terminates them all shortly after starting.

author
#6 lore raw

PATCH 1/1 if-mtu-change.sh: Lower CHANGE_INTERVAL to 1

Quoted 7 lines · Petr Vorel
Quoted 5 lines

Hi Alexey,

Quoted 3 lines

Hi Petr,

Do you mean "sleep $NS_DURATION" in mcast-pktfld02.sh is not needed

Quoted 8 lines
Quoted 3 lines

Yes: sleep $NS_DURATION in mcast-pktfld02.sh and other mcast*.sh tests in:
https://urldefense.com/v3/__https://patchwork.ozlabs.org/project/ltp/patch/[email protected]/__;!!GqivPVa7Brio!P0xbdqXlRQX7GPsKJmcPapdEOX5VV4OnkCKv-P9OqIWNcTvY9a9KXhOSjdJ1dKco5YbG$
https://urldefense.com/v3/__https://patchwork.ozlabs.org/project/ltp/patch/[email protected]/__;!!GqivPVa7Brio!P0xbdqXlRQX7GPsKJmcPapdEOX5VV4OnkCKv-P9OqIWNcTvY9a9KXhOSjdJ1dKjzIVpO$

It's needed because the commands run in parallel in the background during $NS_DURATION. Without it, the test terminates them all shortly after starting.

Thanks for info. Yes, that's right. I got confused, because on netns it's (at least on the default setup), that it passes even without sleep. Maybe we could remove it for netns.

Kind regards,
Petr