search for: zercopy

The following includes vhost-net fixes - both in the experimental zero copy mode. Please pull for 3.1. Thanks!

The following includes vhost-net fixes - both in the experimental zero copy mode. Please pull for 3.1. Thanks!

These packet counters are used to drive the zercopy selection heuristic so nothing too bad happens if they are off a bit - and they are also reset once in a while. But it's cleaner to clear them when backend is set so that we start in a known state. Signed-off-by: Michael S. Tsirkin <mst at redhat.com> --- drivers/vhost/net.c | 3 +++ 1...

These packet counters are used to drive the zercopy selection heuristic so nothing too bad happens if they are off a bit - and they are also reset once in a while. But it's cleaner to clear them when backend is set so that we start in a known state. Signed-off-by: Michael S. Tsirkin <mst at redhat.com> --- drivers/vhost/net.c | 3 +++ 1...

...ver XDP was used. I was going to dig into it but not yet. > > Right, just to confirm this. This is expected. > > In tuntap, we do native XDP only for small and non zerocopy packets. See > tun_can_build_skb(). The reason is XDP may adjust packet header which is > not supported by zercopy. We can only use XDP generic for zerocopy in > this case. I think I understand when driver XDP can be used. What I'm not sure and was going to narrow down is why zerocopy is mostly not applied. -- Toshiaki Makita

...ver XDP was used. I was going to dig into it but not yet. > > Right, just to confirm this. This is expected. > > In tuntap, we do native XDP only for small and non zerocopy packets. See > tun_can_build_skb(). The reason is XDP may adjust packet header which is > not supported by zercopy. We can only use XDP generic for zerocopy in > this case. I think I understand when driver XDP can be used. What I'm not sure and was going to narrow down is why zerocopy is mostly not applied. -- Toshiaki Makita

...dig into it but not yet. >>> Right, just to confirm this. This is expected. >>> >>> In tuntap, we do native XDP only for small and non zerocopy packets. See >>> tun_can_build_skb(). The reason is XDP may adjust packet header which is >>> not supported by zercopy. We can only use XDP generic for zerocopy in >>> this case. >> I think I understand when driver XDP can be used. What I'm not sure and >> was going to narrow down is why zerocopy is mostly not applied. >> > > I see, any touch to the zerocopy packet (clone, head...

...dig into it but not yet. >>> Right, just to confirm this. This is expected. >>> >>> In tuntap, we do native XDP only for small and non zerocopy packets. See >>> tun_can_build_skb(). The reason is XDP may adjust packet header which is >>> not supported by zercopy. We can only use XDP generic for zerocopy in >>> this case. >> I think I understand when driver XDP can be used. What I'm not sure and >> was going to narrow down is why zerocopy is mostly not applied. >> > > I see, any touch to the zerocopy packet (clone, head...

...nd for simplicity, batched > > > processing were simply disabled by only fetching and processing one > > > descriptor at a time, this could be optimized in the future. > > > > > > XDP_DROP (without touching skb) on tun (with Moongen in guest) with > > > zercopy disabled: > > > > > > Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz: > > > Before: 3.20Mpps > > > After: 3.90Mpps (+22%) > > > > > > No differences were seen with zerocopy enabled. > > > > > > Signed-off-by: Jason Wang <jasow...

...nd for simplicity, batched > > > processing were simply disabled by only fetching and processing one > > > descriptor at a time, this could be optimized in the future. > > > > > > XDP_DROP (without touching skb) on tun (with Moongen in guest) with > > > zercopy disabled: > > > > > > Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz: > > > Before: 3.20Mpps > > > After: 3.90Mpps (+22%) > > > > > > No differences were seen with zerocopy enabled. > > > > > > Signed-off-by: Jason Wang <jasow...

...ing > more batching on top. For zerocopy case and for simplicity, batched > processing were simply disabled by only fetching and processing one > descriptor at a time, this could be optimized in the future. > > XDP_DROP (without touching skb) on tun (with Moongen in guest) with > zercopy disabled: > > Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz: > Before: 3.20Mpps > After: 3.90Mpps (+22%) > > No differences were seen with zerocopy enabled. > > Signed-off-by: Jason Wang <jasowang at redhat.com> So where is the speedup coming from? I'd guess the ri...

...ing > more batching on top. For zerocopy case and for simplicity, batched > processing were simply disabled by only fetching and processing one > descriptor at a time, this could be optimized in the future. > > XDP_DROP (without touching skb) on tun (with Moongen in guest) with > zercopy disabled: > > Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz: > Before: 3.20Mpps > After: 3.90Mpps (+22%) > > No differences were seen with zerocopy enabled. > > Signed-off-by: Jason Wang <jasowang at redhat.com> So where is the speedup coming from? I'd guess the ri...

Hi: This series implement batch updating of used ring for TX. This help to reduce the cache contention on used ring. The idea is first split datacopy path from zerocopy, and do only batching for datacopy. This is because zercopy had already supported its own batching. TX PPS was increased 25.8% and Netperf TCP does not show obvious differences. The split of datapath will also be helpful for future implementation like in order completion. Please review. Thanks Jason Wang (9): vhost_net: drop unnecessary parameter v...

Hi: This series implement batch updating of used ring for TX. This help to reduce the cache contention on used ring. The idea is first split datacopy path from zerocopy, and do only batching for datacopy. This is because zercopy had already supported its own batching. TX PPS was increased 25.8% and Netperf TCP does not show obvious differences. The split of datapath will also be helpful for future implementation like in order completion. Please review. Thanks Jason Wang (9): vhost_net: drop unnecessary parameter v...

...expected. >>>>> >>>>> In tuntap, we do native XDP only for small and non zerocopy >>>>> packets. See >>>>> tun_can_build_skb(). The reason is XDP may adjust packet header >>>>> which is >>>>> not supported by zercopy. We can only use XDP generic for zerocopy in >>>>> this case. >>>> I think I understand when driver XDP can be used. What I'm not sure and >>>> was going to narrow down is why zerocopy is mostly not applied. >>>> >>> I see, any touch t...

...ing on top. For zerocopy case and for simplicity, batched >> processing were simply disabled by only fetching and processing one >> descriptor at a time, this could be optimized in the future. >> >> XDP_DROP (without touching skb) on tun (with Moongen in guest) with >> zercopy disabled: >> >> Intel(R) Xeon(R) CPU E5-2650 0 @ 2.00GHz: >> Before: 3.20Mpps >> After: 3.90Mpps (+22%) >> >> No differences were seen with zerocopy enabled. >> >> Signed-off-by: Jason Wang <jasowang at redhat.com> > So where is the speedup...

...20, 2018 at 08:15:12AM +0800, Jason Wang wrote: > Hi: > > This series implement batch updating of used ring for TX. This help to > reduce the cache contention on used ring. The idea is first split > datacopy path from zerocopy, and do only batching for datacopy. This > is because zercopy had already supported its own batching. > > TX PPS was increased 25.8% and Netperf TCP does not show obvious > differences. > > The split of datapath will also be helpful for future implementation > like in order completion. > > Please review. > > Thanks Acked-by:...

...sed. I was going to dig into it but not yet. >> Right, just to confirm this. This is expected. >> >> In tuntap, we do native XDP only for small and non zerocopy packets. See >> tun_can_build_skb(). The reason is XDP may adjust packet header which is >> not supported by zercopy. We can only use XDP generic for zerocopy in >> this case. > I think I understand when driver XDP can be used. What I'm not sure and > was going to narrow down is why zerocopy is mostly not applied. > I see, any touch to the zerocopy packet (clone, header expansion or segmentat...

...yet. >>>> Right, just to confirm this. This is expected. >>>> >>>> In tuntap, we do native XDP only for small and non zerocopy packets. See >>>> tun_can_build_skb(). The reason is XDP may adjust packet header which is >>>> not supported by zercopy. We can only use XDP generic for zerocopy in >>>> this case. >>> I think I understand when driver XDP can be used. What I'm not sure and >>> was going to narrow down is why zerocopy is mostly not applied. >>> >> I see, any touch to the zerocopy packe...

...big patchset. Should be done by Tuesday. -next material anyway. > More optimizations (e.g IN_ORDER) is on the road. > > Please review. > > [1] https://docs.oasis-open.org/virtio/virtio/v1.1/csprd01/virtio-v1.1-csprd01.html#x1-610007 > > This version were tested with: > - zercopy/datacopy > - mergeable buffer on/off > - TCP stream & virtio-user > > Changes from V2: > - rebase on top of vhost metadata accelreation series > - introduce shadow used ring API > - new SET_VRING_BASE/GET_VRING_BASE that takes care about warp counter > and index for b...