Latency: the #1 metric of your cloud
Boyan Krosnov ● Chief of Product ● Cloud architect ● "External technology" ● Previously service providers, networks, packet processing (SDN) ● Algorithms & data structures 2
StorPool ● Fast and efficient software-defined storage system ● Used by Private clouds, MSPs, Cloud service providers, Mobile/Web Apps ● Started in 2011 to solve storage for cloud service providers ● Clean slate design - scale-out, API-controlled, end-to-end data integrity, CoW 3
StorPool ● Most deployments are with KVM ○ some Xen, some VMWare, some Hyper-V ○ some bare metal / dedicated server ● Deep integrations into OpenStack , Kubernetes, CloudStack, OpenNebula, OnApp 4
5 ops per second Latency
task core core core core 6
task task core core core core 7
task task task task core core core core 8
task task task task task task core core core core 9
10 ops per second elastic mode s a t u Latency r a t i o n p o i n t c o n g e s t e d m o d e
ops per second best service Latency 11
lowest cost per delivered resource ops per second best service Latency 12
only pain lowest cost per delivered resource ops per second system throughput best service Latency 13
only pain lowest cost per delivered resource benchmarks ops per second best service Latency 14
15
8 vCPUs, 16 GB RAM vdisk with 20k IOPS "fast" volume with approx 0.15 ms latency (QD 1) "slow" volume with 2 ms latency (QD 1) added with dm-delay in host Both volumes are on the same pool Both volumes measure 20k IOPS flat 16
17 ops per second Latency 20k IOPS
18 ops per second 2ms Latency 20k IOPS
19 ops per second 0.15ms Latency 20k IOPS
8 vCPUs, 16 GB RAM vdisk with 20k IOPS pgbench --client=16 --jobs=8 \ --progress=1 --time=10 pgbench09 database size: 64 GB (4x RAM) https://wiki.postgresql.org/wiki/Pgbenchtesting 20
8 vCPUs, 16 GB RAM vdisk with 20k IOPS 2ms storage latency 1600 TPS @ 8 ms in pgbench -> if we ask for 2500 TPS -> pile up 21
8 vCPUs, 16 GB RAM vdisk with 20k IOPS 0.15ms storage latency 3600 TPS @ 3 ms in pgbench -> if we ask for 2500 TPS - no problem 22
random read/write 50/50 QD 1, avg. latency DigitalOcean (Ceph) 1.75 ms OVH (Ceph) 1.53 ms Tencent (?) 0.76 ms AWS gp2 10k 0.29 ms Alibaba (?) 0.18 ms StorPool BCP 0.17 ms 23
random read/write 50/50 QD 1, avg. latency DigitalOcean (Ceph) 1.75 ms OVH (Ceph) 1.53 ms Tencent (?) 0.76 ms AWS gp2 10k 0.29 ms Alibaba (?) 0.18 ms StorPool BCP 0.17 ms 24
8 vCPUs, 16 GB RAM, dedicated if possible database size = 4x RAM https://wiki.postgresql.org/wiki/Pgbenchtesting 25
Q&A 26
Thank you! 27
Boyan Krosnov StorPool Storage bk@storpool.com www.storpool.com @storpool 28
Recommend
More recommend