Towards Predictable Datacenter Networks

Transcription

1 Towards Predictable Datacenter Networks Hitesh Ballani, Paolo Costa, Thomas Karagiannis and Ant Rowstron Microsoft Research, Cambridge

2 This talk is about Guaranteeing network performance for tenants in multi-tenant datacenters Multi-tenant datacenters Datacenters with multiple (possibly competing) tenants Private datacenters Run by organizations like Facebook, Intel, etc. Tenants: Product groups and applications Cloud datacenters Amazon EC2, Microsoft Azure, Rackspace, etc. Tenants: Users renting virtual machines

3 Tenant Cloud datacenters 101 Simple interface: Tenants ask for a set of VMs Amazon EC2 Interface Request VMs Charging is per-vm, per-hour Amazon EC2 small instances: $0.085/hour No (intra-cloud) network cost Network performance is not guaranteed Bandwidth between a tenant s VMs depends on their placement, network load, protocols used, etc.

4 Performance variability in Up the to 5x wild variability Study Study Provider Duration A [Giurgui 10] Amazon EC2 n/a B [Schad 10] Amazon EC2 31 days C/D/E [Li 10] (Azure, EC2, Rackspace) 1 day F/G [Yu 10] Amazon EC2 1 day H [Mangot 09] Amazon EC2 1 day

5 Network performance can vary... so what? Tenant Tenant Map Reduce Job Data analytics on an isolated cluster Enterprise Results Data analytics in a multi-tenant datacenter Completion Time 4 hours Unpredictability of application performance and tenant costs is a key hindrance to cloud adoption Key Map Contributor: Reduce Network Results performance variation Job Datacenter Completion Time hours Variable tenant costs Expected cost (based on 4 hour completion time) = $100 Actual cost = $

6 Predictable datacenter networks Extend the tenant-provider interface to account for the network Tenant Request Request Virtual Network # of VMs and network demands # of VMs and network demands Contributions- VM 1 VM 2 VM N Virtual network abstractions To capture tenant network demands Oktopus: Proof of concept system Implements virtual networks in multi-tenant datacenters Can be incrementally deployed today! Key Idea: Tenants are offered a virtual network with bandwidth guarantees This decouples tenant performance from provider infrastructure

7 Introduction Talk Outline Virtual network abstractions Oktopus Allocating virtual networks Enforcing virtual networks Evaluation

8 Abstraction 1: Virtual Cluster (VC) Motivation: In enterprises, tenants run applications on dedicated Ethernet clusters Virtual Switch Total bandwidth = N * B VM 1 VM 2 VM N Request <N, B> N VMs. Each VM can send and receive at B Mbps Tenants get a network with no oversubscription Suitable for data-intensive apps. (MapReduce, BLAST) Moderate provider flexibility

9 Abstraction 2: Virtual Oversubscribed Cluster (VOC) VMs can send traffic to group members at B Mbps Group Virtual Switch Root Virtual Switch Total bandwidth at root = N * B / O Total bandwidth at VMs = N * B VM 1 Group 1 VM N S VM 1 Group 2 VM S. VM 1 VM S Group N/S Motivation: Many Request applications <N, B, moving S, O> to the cloud have N VOC VMs capitalizes in groups of size S. Oversubscription factor O. localized on communication tenant communication patterns patterns Applications No oversubscription Suitable are composed for typical for of intra-group applications groups with (though communication more traffic not all) within (captures Intra-group the Improved sparseness groups communication than of provider across inter-group is the groups flexibility common communication) case! Oversubscription factor O for inter-group communication

10 Introduction Talk Outline Virtual network abstractions Oktopus Allocating virtual networks Enforcing virtual networks Evaluation

11 Oktopus Offers virtual networks to tenants in datacenters Two main components Management plane: Allocation of tenant requests Allocates tenant requests to physical infrastructure Accounts for tenant network bandwidth requirements Data plane: Enforcement of virtual networks Enforces tenant bandwidth requirements Achieved through rate limiting at end hosts

12 Allocating Virtual Clusters Request : <3 VMs, 100 Mbps> Max Sending Rate = 2*100 = 200 VM What for an bandwidth existing Max Receive tenant Rate = needs to be 1*100 = 100 reserved for the B/W tenant needed on this on link? = Min (200, 100) = 100Mbps For a virtual cluster <N,B>, bandwidth needed on a link that How Datacenter to Tenant find Physical a valid Request allocation? Topology connects m VMs to the remaining (N-m) VMs is = Min (m, N-m) * B An allocation Link divides of tenant virtual VMs tree to into physical two parts machines Allocate 4 physical a tenant machines, asking for 2 3 VM VMs slots arranged per machine a virtual cluster Tenant Consider with traffic all traffic 100 Mbps traverses from each, the the i.e. highlighted left to right <3 VMs, 100Mbps> links part Bandwidth needed <= Link s Residual Bandwidth For a valid allocation:

13 Allocation Algorithm Request : <3 VMs, 100 Mbps> How Solution many VMs can At most be allocated 1 VM for to this tenant machine? can be 2 VMs allocated here 2 VMs 2 VMs 1 VM 1 VM Constraints for Allocation # of VMs (m) is that fast can and be allocated efficient to the machine- 3 VMs Greedy Key allocation intuition Packing algorithm Traverse Validity up the conditions VMs together hierarchy can motivated and be determine used to by determine the fact that the lowest the level at number datacenter of VMs networks which that all can are 3 VMs be can allocated typically oversubscribed be allocated to any level of the Allocation datacenter; can be extended machines, for goals racks like failure and so resiliency, on etc. 1. VMs can only be allocated to empty slots m <= VMs are requested m <= 3 3. Enough b/w on outbound link min (m, 3-m)*100 <= 200

14 Introduction Talk Outline Virtual network abstractions Oktopus Allocating virtual networks Enforcing virtual networks Evaluation

15 Enforcement in Oktopus: Key highlights Oktopus enforces virtual networks at end hosts Use egress rate limiters at end hosts Implement on hypervisor/vmm Oktopus can be deployed today No changes to tenant applications No network support Tenants without virtual networks can be supported Good for incremental roll out

16 Introduction Talk Outline Virtual network abstractions Oktopus Allocating virtual networks Enforcing virtual networks Evaluation

17 Datacenter Simulator Flow-based simulator 16,000 servers and 4 VMs/server 64,000 VMs Three-tier network topology (10:1 oversubscription) Tenants submit requests for VMs and execute jobs Job: VMs process and shuffle data between each other Baseline: representative of today s setup Tenants simply ask for VMs VMs are allocated in a locality-aware fashion Virtual network request Tenants ask for Virtual Cluster (VC) or Virtual Oversubscribed Cluster (VOC)

18 Worse Private datacenters VC is Virtual Cluster VOC-10 is Virtual Oversubscribed Cluster with oversubscription=10 Execute a batch of 10,000 tenant jobs Jobs vary in network intensiveness (bandwidth at which a job can generate data) Better Virtual networks improve completion time VC: 50% of Baseline VOC-10: 31% of Baseline Jobs become more network intensive

19 Cloud Datacenters Amazon EC2 s reported target utilization Worse Tenant job requests arrive over time Jobs are rejected if they cannot be accommodated on arrival (representative of cloud datacenters) Better Job requests arrive faster Rejected Requests Baseline: 31% VC: 15% VOC-10: 5%

20 Tenant Costs What should tenants pay to ensure provider revenue neutrality, i.e. provider revenue remains the same with all approaches Based on today s EC2 prices, i.e. $0.085/hour for each VM Provider revenue increases while tenants pay less At 70% target utilization, provider revenue increases by 20% and median tenant cost reduces by 42%

21 Oktopus Deployment Implementation scales well and imposes low overhead Allocation of virtual networks is fast In a datacenter with 10 5 machines, median allocation time is 0.35ms Enforcement of virtual networks is cheap Use Traffic Control API to enforce rate limits at end hosts Deployment on testbed with 25 end hosts End hosts arranged in five racks

22 Oktopus Deployment Cross-validation of simulation results Completion time for jobs in the simulator matches that on the testbed

23 Summary Proposal: Offer virtual networks to tenants Virtual network abstractions Resemble physical networks in enterprises Make transition easier for tenants Proof of concept: Oktopus Tenants get guaranteed network performance Sufficient multiplexing for providers Win-win: tenants pay less, providers earn more! How to determine tenant network demands? Ongoing work: Map high-level goals (like desired completion time) to Oktopus abstractions

24 Thank you