act that exec_upcalls calls dpif_print_packet() is a bit weird,
> feels like a layering violation. I'm conflicted though, what do you
> think?
>
>
Not sure why this would be since we call dpif_recv in ofproto-upcall.
> Ethan
>
> On Sun, Jul 20, 2014 at 3:37 PM, Ryan Wils
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
v2: Fix race condition found during perf test
v3: Addressed Daniele's comments
v4: Addressed Ben's style comments, added packet batching
v5: Rebase
v
eated with a new
type. I tested out the fix and it works with my patch as well.
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 4, 2014, at 4:18 PM, Joe Stringer wrote:
> On 3 May 2014 02:56, Ben Pfaf
Hey Joe,
This looks good to me and fixes the root cause which is even better. I'm fairly
new to OVS, so my ack likely doesn't mean much :)
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 5, 201
This patch also adds a ref/unref structure for ofgroup, so handler and
revalidator threads can update group / bucket stats via the xlate cache.
See the following thread for more information:
http://openvswitch.org/pipermail/dev/2014-January/036107.html
Signed-off-by: Ryan Wilson
---
lib/ofp
.
This patch also frees netdev with ovsrcu_postpone. This is because if RCU
reader threads take a ref to the netdev, the netdev cannot be deleted and
re-created with different types.
Signed-off-by: Ryan Wilson
---
lib/netdev.c | 37 +--
lib/ovs-rcu.h|2
This patch also adds a ref/unref structure for ofgroup, so handler and
revalidator threads can update group / bucket stats via the xlate cache.
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more com
is no need
for this hash map.
This patch removes this hash map and each packet has its own upcall.
Signed-off-by: Ryan Wilson
---
ofproto/ofproto-dpif-upcall.c | 280 +
1 file changed, 89 insertions(+), 191 deletions(-)
diff --git a/ofproto/ofproto-dpif
y commit.
After re-reading my commit, I totally agree. I'll resubmit the patch split into
2 with a more clear description.
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 9, 2014, at 9:59 AM, Alex Wan
.
Signed-off-by: Ryan Wilson
---
v2: Split 1 commit into 3:
- netdev: Free netdev with ovsrcu_postpone.
- ovs-rcu: Add OVSRCU_TYPE_INITIALIZER which initializes OVSRCU_TYPE variables
to NULL.
- ofproto-dpif-xlate: Implement RCU locking in ofproto-dpif-xlate.
---
ofproto/ofproto-dpif
---
lib/ovs-rcu.h |2 ++
1 file changed, 2 insertions(+)
diff --git a/lib/ovs-rcu.h b/lib/ovs-rcu.h
index 4b451b2..8327517 100644
--- a/lib/ovs-rcu.h
+++ b/lib/ovs-rcu.h
@@ -129,6 +129,7 @@
*/
#if __GNUC__
#define OVSRCU_TYPE(TYPE) struct { ATOMIC(TYPE) p; }
+#define OVSRCU_TYPE_INITIALIZ
properties
while the main thread is not blocked from making netdev configuration changes.
Note that any RCU reader threads reading netdev should not call netdev_ref or
netdev_close.
Signed-off-by: Ryan Wilson
---
lib/netdev.c | 37 +++--
1 file changed, 23
r the memory
overhead of allocating a hash map and this would outweigh the negative effect
of possibly re-processing a packet from the same flow. Since userspace
forwarding has gotten so much faster, this wouldn't be so bad.
Obviously, I need perf numbers to justify these claims; I will post
When the user changes port type (i.e. changing p0 from type 'internal' to
'gre'), the netdev must first be deleted, then re-created with the new type.
Deleting the netdev requires there exist no more references to the netdev.
However, the xlate cache holds references to netdevs and the cache is onl
this and if this sounds good,
I'll resubmit my RCU locking patch corrected for this change.
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 9, 2014, at 11:48 AM, Alex Wang wrote:
> Great to see this s
nges port type. This ensures that the new netdev can always be created
while handler and revalidator threads can retain references to the old netdev
until they are finished.
Signed-off-by: Ryan Wilson
Acked-by: Ben Pfaff
---
v2: Clarified comments for netdev_remove based on Alex Wa
It appears this patch will be pushed, so I'll send out v3 of the xlate RCU
locking patch.
Cheers,
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 15, 2014, at 11:16 AM, Ryan Wilson wrote:
> H
---
lib/ovs-rcu.h |2 ++
1 file changed, 2 insertions(+)
diff --git a/lib/ovs-rcu.h b/lib/ovs-rcu.h
index 4b451b2..8327517 100644
--- a/lib/ovs-rcu.h
+++ b/lib/ovs-rcu.h
@@ -129,6 +129,7 @@
*/
#if __GNUC__
#define OVSRCU_TYPE(TYPE) struct { ATOMIC(TYPE) p; }
+#define OVSRCU_TYPE_INITIALIZ
.
Signed-off-by: Ryan Wilson
---
v2: Split 1 commit into 3:
- netdev: Free netdev with ovsrcu_postpone.
- ovs-rcu: Add OVSRCU_TYPE_INITIALIZER which initializes OVSRCU_TYPE variables
to NULL.
- ofproto-dpif-xlate: Implement RCU locking in ofproto-dpif-xlate.
v3: Separate netdev patch
.
Signed-off-by: Ryan Wilson
Acked-by: Alex Wang
---
v2: Split 1 commit into 3:
- netdev: Free netdev with ovsrcu_postpone.
- ovs-rcu: Add OVSRCU_TYPE_INITIALIZER which initializes OVSRCU_TYPE variables
to NULL.
- ofproto-dpif-xlate: Implement RCU locking in ofproto-dpif-xlate.
v3
Signed-off-by: Ryan Wilson
Acked-by: Alex Wang
---
lib/ovs-rcu.h |2 ++
1 file changed, 2 insertions(+)
diff --git a/lib/ovs-rcu.h b/lib/ovs-rcu.h
index 4b451b2..8327517 100644
--- a/lib/ovs-rcu.h
+++ b/lib/ovs-rcu.h
@@ -129,6 +129,7 @@
*/
#if __GNUC__
#define OVSRCU_TYPE(TYPE) struct
Shoot, total brain fart on my part. Just sent out another version of the patch
with the signed off by.
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 19, 2014, at 7:59 AM, Ben Pfaff wrote:
> Looks g
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group reference count, support for group and bucket stats into 2
patches. Commit Andy Zhou's test patch
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group ref/unref, support for group and bucket stats, and Andy Zhou's
tests into 3 patches.
---
lib/
From: Andy Zhou
Signed-off-by: Andy Zhou
Signed-off-by: Ryan Wilson
---
tests/ofproto-dpif.at | 44
1 file changed, 44 insertions(+)
diff --git a/tests/ofproto-dpif.at b/tests/ofproto-dpif.at
index fad8f2c..b87f735 100644
--- a/tests/ofproto
duplicate flow additions.
With hash map in upcall:
NUM RESULTS: 23944
MEAN: 150.843937
MEDIAN: 150.66
Without hash map in up call:
NUM RESULTS: 24736
MEAN: 147.618262
MEDIAN: 147.300000
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Offi
real 1m28.357s
Deleting ports: real1m33.277s
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1511 Office
916.588.7783 Mobile
On May 19, 2014, at 8:56 AM, Ben Pfaff wrote:
> On Fri, May 16, 2014 at 06:59:02AM -0700, Ryan Wils
Sorry Gurucharan, totally forgot to answer your question!
After interspersing these tests with random calls to reload the kernel module,
it doesn't appear to affect time in any significant way.
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alt
/s):
RCU:
Mean: 84.591932
Median: 83.405000
Master:
Mean: 78.528627
Median: 70.55
Its not huge, but if we add more ovs-vsctl commands, I'd imagine we'd see more
improvement. Not sure if this is a valid use case, but these are my findings so
far.
Ryan Wilson
Member of Technical St
,in_port(4),skb_mark(0),eth(src=a0:36:9f:33:3a:c0,dst=a2:2e:02:45:b6:14),eth_type(0x0800),ipv4(src=1.1.1.110,dst=1.1.1.30,proto=6,tos=0,ttl=64,frag=no),tcp(src=54622,dst=41606),tcp_flags(0x010)
Ryan Wilson
Member of Technical Staff
wr...@vmware.com
3401 Hillview Avenue, Palo Alto, CA
650.427.1
is no need
for this hash map.
This patch removes this hash map and each packet has its own upcall.
Signed-off-by: Ryan Wilson
---
v2: Addressed Alex's comments
---
ofproto/ofproto-dpif-upcall.c | 283 +
1 file changed, 91 insertions(+), 192 dele
cleared by a
revalidator thread.
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group reference count, support for group and bucket stats into 2
patches. Commit
cleared by a
revalidator thread.
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group reference count, support for group and bucket stats into 2
patches. Commit
o the
group. This also eliminates the need for ofgroup's lock as all properties of
ofgroup are read-only.
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group
o the
group. This also eliminates the need for ofgroup's lock as all properties of
ofgroup are read-only.
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group
s the need for ofgroup's lock.
Signed-off-by: Ryan Wilson
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group reference count, support for group and bucket stats into 2
patches. Commit An
Signed-off-by: Ryan Wilson
Acked-by: Andy Zhou
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group ref/unref, support for group and bucket stats, and Andy Zhou's
tests into 3 patche
Signed-off-by: Ryan Wilson
Acked-by: Andy Zhou
---
v2: Fixed bug with group stats all buckets, cleaned up ofgroup unref code,
added Andy Zhou's test for more complete test coverage
v3: Split group ref/unref, support for group and bucket stats, and Andy Zhou's
tests into 3 patche
Signed-off-by: Ryan Wilson
---
TODO | 17 -
1 file changed, 17 deletions(-)
diff --git a/TODO b/TODO
index dabe49c..e11089a 100644
--- a/TODO
+++ b/TODO
@@ -52,23 +52,6 @@ These changes might have backward-compatibility
implications; one
would have to test the behavior of
.
Signed-off-by: Ryan Wilson
Acked-by: Alex Wang
Acked-by: Ethan Jackson
---
v2: Split 1 commit into 3:
- netdev: Free netdev with ovsrcu_postpone.
- ovs-rcu: Add OVSRCU_TYPE_INITIALIZER which initializes OVSRCU_TYPE variables
to NULL.
- ofproto-dpif-xlate: Implement RCU locking in
freed. This causes a
race condition with the main thread which calls route_table_run
periodically to check for routing table updates.
This patch makes the route-table module thread-safe via a mutex.
Bug #1258532
Signed-off-by: Ryan Wilson
---
lib/route-table.c | 19 +++
1 file
freed. This causes a
race condition with the main thread which calls route_table_run
periodically to check for routing table updates.
This patch makes the route-table module thread-safe via a mutex.
Bug #1258532
Signed-off-by: Ryan Wilson
Acked-by: Ethan Jackson
---
lib/route-table.c | 23
. Since the
rtbsd_notifier_run() is always run by the main thread often,
receiving stale notifications from the notifier is unlikely.
Bug #1258532
Signed-off-by: Ryan Wilson
---
lib/rtbsd.c | 39 +--
1 file changed, 29 insertions(+), 10 deletions(-)
diff
last transaction was
not successful.
Signed-off-by: Ryan Wilson
---
vswitchd/bridge.c | 37 ++---
1 file changed, 22 insertions(+), 15 deletions(-)
diff --git a/vswitchd/bridge.c b/vswitchd/bridge.c
index 9764c1f..9137dcd 100644
--- a/vswitchd/bridge.c
+++ b
Signed-off-by: Ryan Wilson
---
lib/ovsdb-idl.c | 28 +---
1 file changed, 25 insertions(+), 3 deletions(-)
diff --git a/lib/ovsdb-idl.c b/lib/ovsdb-idl.c
index 7556b7f..d6b9326 100644
--- a/lib/ovsdb-idl.c
+++ b/lib/ovsdb-idl.c
@@ -23,6 +23,7 @@
#include
#include
;s version >= 3.3.
Otherwise, the timeval module calls out to the librt C shared library and uses
the clock_gettime function with CLOCK_MONOTONIC_*.
Note this is only enabled on Linux-based platforms.
Bug #1189434
Signed-off-by: Ryan Wilson
---
python/ovs/time
change again. This could
leave the database in an incorrect state for a long period of time.
This patch always sends status changes to the database if the last
transaction was not successful.
Signed-off-by: Ryan Wilson
---
v2: Addressed Alex's comments, edited commit message to be more
acc
. Since the
rtbsd_notifier_run() is always run by the main thread often,
receiving stale notifications from the notifier is unlikely.
Bug #1258532
Signed-off-by: Ryan Wilson
Acked-by: YAMAMOTO Takashi
---
v2: Fix OVS_EXCLUDES/OVS_EXCLUDED typo
---
lib/rtbsd.c | 39
change again. This could
leave the database in an incorrect state for a long period of time.
This patch always sends status changes to the database if the last
transaction was not successful.
Signed-off-by: Ryan Wilson
---
v2: Addressed Alex's comments, edited commit message to be more
acc
change again. This could
leave the database in an incorrect state for a long period of time.
This patch always sends status changes to the database if the last
transaction was not successful.
Signed-off-by: Ryan Wilson
---
v2: Addressed Alex's comments, edited commit message to be more
acc
change again. This could
leave the database in an incorrect state for a long period of time.
This patch always sends status changes to the database if the last
transaction was not successful.
Bug #1256577
Signed-off-by: Ryan Wilson
---
v2: Addressed Alex's comments, edited commit message
;s
version >= 3.3. Otherwise, the timeval module calls out to the
librt C shared library and uses the clock_gettime function with
CLOCK_MONOTONIC.
Note this is only enabled on Linux-based platforms. This has been
tested on Ubuntu 12.04 and Redhat 6.4.
Bug #1189434
Signed-off-by: Ryan Wilson
---
hat 6.4.
Bug #1189434
Signed-off-by: Ryan Wilson
---
v2: Pre-load librt.so.1 library instead of loading at every function
call. Use only CLOCK_MONOTONIC for consistency with OVS C library.
v3: Edit commit message, remove if-linux check
---
python/ovs/timeval.py | 34
Older versions of Python do not have ctypes as a default installed
package. This patch puts the 'import ctypes' statement inside a try
statement.
This fixes a bug introduced by commit 8396f (timeval: Use monotonic
time in OVS Python timeval library).
Signed-off-by: Ryan Wilson
---
When a bridge of datatype type netdev receives a packet, it copies the
packet from the NIC to a buffer in userspace. Currently, when making
an upcall, the packet is again copied to the upcall's buffer. However,
this extra copy is not necessary when the datapath exists in userspace
as the upcall can
ch ensures Linux, NetBSD and FreeBSD platforms use
clock_gettime with their corresponding correct values and
definitions. All other platforms use time.time().
Signed-off-by: Ryan Wilson
---
python/ovs/timeval.py | 21 +
1 file changed, 17 insertions(+), 4 deletions(-)
di
ch ensures Linux, NetBSD and FreeBSD platforms use
clock_gettime with their corresponding correct values and
definitions. All other platforms use time.time().
Signed-off-by: Ryan Wilson
---
v2: Added Yamamoto's diff to fix NetBSD case
---
python/ovs/timeval.py | 34 ++
pcall can directly access the packet data.
This patch eliminates this extra copy of the packet data in most cases.
In cases where the packet may still be used later by callers of
dp_netdev_execute_actions, making a copy of the packet data is still
necessary.
Signed-off-by: Ryan Wilson
---
v2: Addr
pcall can directly access the packet data.
This patch eliminates this extra copy of the packet data in most cases.
In cases where the packet may still be used later by callers of
dp_netdev_execute_actions, making a copy of the packet data is still
necessary.
Signed-off-by: Ryan Wilson
Acked-by:
ing of the allocated buffer from 'struct
rte_mbuf'. This allows ofpbufs to share packet data when using DPDK.
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c |3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index ba41d2e..155c
x27;struct ofpbuf' when using
DPDK. This field holds a pointer to the allocated buffer in the
rte_mempool. Thus, an upcall packet ofpbuf allocated on the stack can
now share data and free memory of a rte_mempool allocated ofpbuf.
Signed-off-by: Ryan Wilson
Acked-by: Jarno Rajahalme
--
x27;struct ofpbuf' when using
DPDK. This field holds a pointer to the allocated buffer in the
rte_mempool. Thus, an upcall packet ofpbuf allocated on the stack can
now share data and free memory of a rte_mempool allocated ofpbuf.
Signed-off-by: Ryan Wilson
Acked-by: Jarno Rajahalme
--
ld to 'struct ofpbuf' when using
DPDK. This field holds a pointer to the allocated DPDK buffer in the
rte_mempool. Thus, an upcall packet ofpbuf allocated on the stack can
now share data and free memory of a rte_mempool allocated ofpbuf.
Signed-off-by: Ryan Wilson
Acked-by: Jarno Rajahalme
--
ld to 'struct ofpbuf' when using
DPDK. This field holds a pointer to the allocated DPDK buffer in the
rte_mempool. Thus, an upcall packet ofpbuf allocated on the stack can
now share data and free memory of a rte_mempool allocated ofpbuf.
Signed-off-by: Ryan Wilson
Acked-by: Jarno Rajahalme
Ack
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
lib/dpif-netdev.c | 332 -
lib/dpif-netdev.h |9 ++
lib/dpif.c| 68
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
v2: Fix race condition found during perf test
---
lib/dpif-netdev.c | 327 +++--
lib/dpif-netdev.h |
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
v2: Fix race condition found during perf test
v3: Addressed Daniele's comments
---
lib/dpif-linux.c |3 +
lib/dpif-netdev.c
dev'.
Signed-off-by: Ryan Wilson
---
lib/dpif-linux.c |7 ++--
lib/dpif-netdev.c | 87 +++--
lib/dpif-provider.h |9 +++--
lib/dpif.c| 10 ++---
lib/dpif.h|4 +-
ofpr
e just assert that it is <= thread_->max_flows.
>
> The comment above dpif_flow_dump_next() may also need to be updated.
>
Sure I'll add the max_flows argument back to dpif_flow_dump_next() and add
an assert. I shouldn't need to change the dpif_flow_dump_next() comment
though b
dev'.
Signed-off-by: Ryan Wilson
---
v2: Addressed Joe's comments, split into 2 patches
---
lib/dpif-netdev.c | 82 +
1 file changed, 45 insertions(+), 37 deletions(-)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 805af9a.
This is needed to add flow dump batching to dpif-netdev, specifically
for malloc-ing 'max_flows' key and mask buffers.
Signed-off-by: Ryan Wilson
---
v2: Addressed Joe's comments, split into 2 patches
---
lib/dpif-linux.c |4 ++--
lib/dpif-netdev.c |
dev'.
Signed-off-by: Ryan Wilson
---
v2: Addressed Joe's comments, split into 2 patches
v3: Optimized patch by allocating memory on stack, reducing time
spent in locked state.
---
lib/dpif-netdev.c | 106 +
1 file changed, 59 insertions(+
dev'.
Signed-off-by: Ryan Wilson
---
v2: Addressed Joe's comments, split into 2 patches
v3: Optimized patch by allocating memory in 1 malloc call instead of 3,
reducing time spent in locked state. Need to only use 1 patch now.
v4: Minor fixes.
title_init().
Signed-off-by: Daniele Di Proietto
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c | 10 --
1 file changed, 8 insertions(+), 2 deletions(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index fbdb6b3..fb8d953 100644
--- a/lib/netdev-dpdk.c
+++ b/lib/netdev-dpdk.c
Just one comment below. Otherwise, all other issues will be addressed in
the next patch.
Thanks!
On Tue, Jun 24, 2014 at 8:44 AM, Ben Pfaff wrote:
> On Wed, Jun 18, 2014 at 11:07:12AM -0700, Ryan Wilson wrote:
> > Typically, kernel datapath threads send upcalls to userspace where
&g
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
v2: Fix race condition found during perf test
v3: Addressed Daniele's comments
v4: Addressed Ben's style comments, added packet batching
-
>
>
> Thanks Ryan!
>
> I folded in this change to ensure that the caller could specify any
> 'max_flows' that it likes. It doesn't make sense for the caller to
> have to be aware of the implementation's limitations:
>
>
I was debating between either using an assert or using the min of max_flows
an
miniflow_destroy() needs to be called after using miniflow_init().
Otherwise, if the miniflow mallocs data, then a memory leak may
occur.
Found by inspection.
Signed-off-by: Ryan Wilson
---
lib/dpif-netdev.c |1 +
1 file changed, 1 insertion(+)
diff --git a/lib/dpif-netdev.c b/lib/dpif
The current timestamp should be set every time the queue is flushed.
Thus, if DRAIN_TSC timer cycles have passed since the last timestamp,
the send queue should be flushed again.
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c |8 ++--
1 file changed, 2 insertions(+), 6 deletions
This patch refactors dpdk_queue_flush() to reuse code in
dpdk_queue_pkts().
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c | 37 -
1 file changed, 16 insertions(+), 21 deletions(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index 72add87..084e8cd
would not be sent and leak memory.
This patch also adds OVS_UNLIKELY annotations.
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c | 25 ++---
1 file changed, 14 insertions(+), 11 deletions(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index 6e1d293..03f1e02 100644
The current timestamp should be set every time the queue is flushed.
Thus, if DRAIN_TSC timer cycles have passed since the last timestamp,
the send queue should be flushed again.
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c |8 ++--
1 file changed, 2 insertions(+), 6 deletions
Per style requirements, functions local to a module should not be
inlined.
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c |2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index dfa7d77..0aee14e 100644
--- a/lib/netdev-dpdk.c
+++ b
Since dropped packets due to large packet size or lack of memory
are unlikely, it is best to add OVS_UNLIKELY annotations to these
conditions.
With DPDK fast path forwarding test, this increased throughtput
from 4.12 to 4.16 million packets per second.
Signed-off-by: Ryan Wilson
---
lib/netdev
when
multiple packets are dropped in dpdk_do_tx_copy().
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c | 21 -
1 file changed, 12 insertions(+), 9 deletions(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index 6e1d293..dfa7d77 100644
--- a/lib/netdev-dpdk.c
+++ b/lib/n
This patch refactors dpdk_queue_flush() to reuse code in
dpdk_queue_pkts().
Signed-off-by: Ryan Wilson
---
lib/netdev-dpdk.c | 37 -
1 file changed, 16 insertions(+), 21 deletions(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index 72add87..084e8cd
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
v2: Fix race condition found during perf test
v3: Addressed Daniele's comments
v4: Addressed Ben's style comments, added packet batching
v5: Rebase
-
This fixes valgrind errors.
Signed-off-by: Ryan Wilson
---
ovsdb/ovsdb-server.c | 25 -
1 file changed, 16 insertions(+), 9 deletions(-)
diff --git a/ovsdb/ovsdb-server.c b/ovsdb/ovsdb-server.c
index a85a672..0a1be0f 100644
--- a/ovsdb/ovsdb-server.c
+++ b/ovsdb/ovsdb
upcall functions, eliminating the need for handler threads
for datapaths of type 'netdev'.
Signed-off-by: Ryan Wilson
---
v2: Fix race condition found during perf test
v3: Addressed Daniele's comments
v4: Addressed Ben's style comments, added packet batching
v5: Rebase
v
in ovsdb's Interface table likely
indicates that ovs-vswitchd has been restarted or killed.
Signed-off-by: Ryan Wilson
---
diff --git a/lib/netdev-provider.h b/lib/netdev-provider.h
index 9d8e67f..ea2e59c 100644
--- a/lib/netdev-provider.h
+++ b/lib/netdev-provider.h
@@ -40,6 +
be tied to netdevs to take account
of the case when a tunnel is removed and re-added for whatever reason.
Does that make sense? If so, an explanation of the sort should be
added to the commit message.
Ethan
>
> On Apr 22, 2014 4:09 PM, "Ryan Wilson 76511" < wr...@vmware.co
ve in
iface_destroy__ (which Alex recommended).
- Original Message -
From: "Ben Pfaff"
To: "Ryan Wilson"
Cc: dev@openvswitch.org
Sent: Friday, May 16, 2014 7:52:19 AM
Subject: Re: [ovs-dev] [PATCH] netdev: Remove netdev from global shash when the
user is changing interface co
We should use Andy's patch for the tests since I didn't write the tests:
http://patchwork.openvswitch.org/patch/3626/
I'll split the group RCU and the crediting stats parts of this commit on my
next version.
Ryan
- Original Message -
From: "Alex Wang"
e, meaning 1000 copies
of the configuration in memory. This should stress memory usage. I ran a test
in my previous email that details this behavior more specifically, but I know
I've sent a lot of emails so here's the gist of it.
Ryan
From: Ryan Wilson mailto:wr...@vmware.com>
e lock annotation may be lacking around group
>functions. It would
>be nice to add them.
I'll add these annotations in my next patch.
Thanks for the review,
Ryan
>
>On Mon, May 19, 2014 at 6:20 AM, Ryan Wilson wrote:
>> Signed-off-by: Ryan Wilson
>>
>> --
o.c that seem to be
missing some annotations, but those seem unrelated to my patch. So it might be
better to do them in a separate patch. Let me know what you think.
Ryan
From: Joe Stringer mailto:joestrin...@nicira.com>>
Date: Tuesday, May 20, 2014 3:33 PM
To: Ryan Wilson mailto:wr...@vm
: Ryan Wilson mailto:wr...@vmware.com>>
Cc: "dev@openvswitch.org<mailto:dev@openvswitch.org>"
mailto:dev@openvswitch.org>>, Ryan Wilson
mailto:wr...@nicira.com>>
Subject: Re: [ovs-dev] [PATCH] ofproto: Remove per-flow miss hash table from
upcall handler.
On Mo
>On Tue, May 20, 2014 at 3:54 PM, Ryan Wilson 76511
>wrote:
>>
>> Thanks for the review, Joe! I added a more clear description in 'struct
>> ofgroup' and the commit message to explain why the refcount is needed.
>That's good. Thanks for the improvement.
uesday, May 20, 2014 7:06 PM
To: Ryan Wilson mailto:wr...@vmware.com>>
Cc: Alex Wang mailto:al...@nicira.com>>,
"dev@openvswitch.org<mailto:dev@openvswitch.org>"
mailto:dev@openvswitch.org>>, Ryan Wilson
mailto:wr...@nicira.com>>
Subject: Re: [ovs-dev] [P
>>
>> 1) Use refs for buckets (this seems unnecessarily heavyweight)
>> 2) Always try bucket_lookup when adding stats to a bucket via the cache.
>> If the bucket is not there in the list, we know its been removed via
>> modify_group() and we simply don't update stats.
>> 3) Use ofgroup->rw_lock to
1 - 100 of 135 matches
Mail list logo
