search for: 40us

When the Virtio queue is full, a work item is scheduled to execute in 1ms that retries adding the request to the queue. This is a large amount of time on the scale on which a virtio-fs device can operate. When using a DPU this is around 40us baseline without going to a remote server (4k, QD=1). This patch queues requests when the Virtio queue is full, and when a completed request is taken off, immediately fills it back up with queued requests. This reduces the 99.9th percentile latencies in our tests by 60x and slightly increases the...

When the Virtio queue is full, a work item is scheduled to execute in 1ms that retries adding the request to the queue. This is a large amount of time on the scale on which a virtio-fs device can operate. When using a DPU this is around 40us baseline without going to a remote server (4k, QD=1). This patch queues requests when the Virtio queue is full, and when a completed request is taken off, immediately fills it back up with queued requests. This reduces the 99.9th percentile latencies in our tests by 60x and slightly increases the...

...io queue is full, a work item is scheduled > > >> to execute in 1ms that retries adding the request to the queue. > > >> This is a large amount of time on the scale on which a > > >> virtio-fs device can operate. When using a DPU this is around > > >> 40us baseline without going to a remote server (4k, QD=1). > > >> This patch queues requests when the Virtio queue is full, > > >> and when a completed request is taken off, immediately fills > > >> it back up with queued requests. > > >> > > >&gt...

...t;> When the Virtio queue is full, a work item is scheduled > >> to execute in 1ms that retries adding the request to the queue. > >> This is a large amount of time on the scale on which a > >> virtio-fs device can operate. When using a DPU this is around > >> 40us baseline without going to a remote server (4k, QD=1). > >> This patch queues requests when the Virtio queue is full, > >> and when a completed request is taken off, immediately fills > >> it back up with queued requests. > >> > >> This reduces the 99.9th...

When the Virtio queue is full, a work item is scheduled to execute in 1ms that retries adding the request to the queue. This is a large amount of time on the scale on which a virtio-fs device can operate. When using a DPU this is around 40us baseline without going to a remote server (4k, QD=1). This patch queues requests when the Virtio queue is full, and when a completed request is taken off, immediately fills it back up with queued requests. This reduces the 99.9th percentile latencies in our tests by 60x and slightly increases the...

...is full, a work item is scheduled >>>>> to execute in 1ms that retries adding the request to the queue. >>>>> This is a large amount of time on the scale on which a >>>>> virtio-fs device can operate. When using a DPU this is around >>>>> 40us baseline without going to a remote server (4k, QD=1). >>>>> This patch queues requests when the Virtio queue is full, >>>>> and when a completed request is taken off, immediately fills >>>>> it back up with queued requests. >>>>> >>&...

...10.10.241.44: icmp_seq=99 ttl=64 time=0.039 ms 64 bytes from 10.10.241.44: icmp_seq=125 ttl=64 time=0.195 ms 64 bytes from 10.10.241.44: icmp_seq=126 ttl=64 time=0.263 ms 64 bytes from 10.10.241.44: icmp_seq=127 ttl=64 time=0.210 ms As you can see, the response time before the migration is around 40us, and after, it''s 200us, which is understandable since the VM is now in another physical host. The problem is the 25 lost packets between the last phase of the migration. Don''t get me wrong: 2.5s is a very good time, but 50 times higher than what it is told to be, isn''t....