> -----Original Message----- > From: Vitaly Kuznetsov [mailto:vkuznets@xxxxxxxxxx] > Sent: Tuesday, April 28, 2015 1:04 > To: KY Srinivasan > Cc: Haiyang Zhang; devel@xxxxxxxxxxxxxxxxxxxxxx; linux- > kernel@xxxxxxxxxxxxxxx; Dexuan Cui > Subject: [PATCH v2 0/6] Drivers: hv: vmbus: fair round robin algorithm for > vmbus_get_outgoing_channel() > > Changes in v2: > - Address Dexuan's review comments: > PATCH 3/6: s,channel,primary_channel; > PATCH 4/6: add a forward declaration instead of moving code around; > PATCH 6/6: fix an off-by-one > - Change the algorithm in PATCH 6/6: > Instead of a simple round robin we first try to find a (sub)channel with > the current_cpu == target_cpu and we fallback to a round robin when we > fail > to find one. > > K. Y., Dexuan, can you please give it a spin in various testing environments > you have? Thanks! > > Original description: > > This series is a continuation of the "Drivers: hv: vmbus: Use a round-robin > algorithm for picking the outgoing channel" work. It is supposed to bring > two > significant changes: > 1) Subchannels for a channel are distributed evenly across all vcpus we have. > Currently we try to distribute all channels (including subchannels) across > all vcpus, this approach doesn't guarantee that the particular channel's > subchannels will be distributed in the same way as we process all offer > requests in some random order. (Patch 05) > 2) Channel picking based on the current vcpu is dropped from > vmbus_get_outgoing_channel() in favor of a fair round robin. (Patch 06) > (this > is not true anymore, see 'Changes'). > > Patches 01 - 04 are cleanup/refactoring. > > Vitaly Kuznetsov (6): > Drivers: hv: vmbus: unify calls to percpu_channel_enq() > Drivers: hv: vmbus: briefly comment num_sc and next_oc > Drivers: hv: vmbus: decrease num_sc on subchannel removal > Drivers: hv: vmbus: move init_vp_index() call to vmbus_process_offer() > Drivers: hv: vmbus: distribute subchannels among all vcpus > Drivers: hv: vmbus: improve selection of an outgoing channel > > drivers/hv/channel_mgmt.c | 127 ++++++++++++++++++++++++++--------- > ----------- > include/linux/hyperv.h | 12 +++-- > 2 files changed, 80 insertions(+), 59 deletions(-) > > -- Patch 1, 2 and 3 are good to me. We'll have to test 4~6 for performance change. Thanks, -- Dexuan _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
