Offering Network Performance Guarantees in Multi-Tenant Datacenters

• US 20130014101A1
• Filed: 07/06/2011
• Published: 01/10/2013
• Est. Priority Date: 07/06/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Ballani, Hitesh, Costa, Paolo, Karagiannis, Thomas, Rowstron, Antony
• Granted Patent: US 8,671,407 B2
• Time in Patent Office: Days
• US Class Current: 718/1
• CPC Class Codes: G06F 2009/45579 I/O management, e.g. providing access to device drivers or storage
• CPC Class Codes: G06F 2209/501 Performance criteria
• CPC Class Codes: G06F 9/45533 Hypervisors; Virtual machine monitors
• CPC Class Codes: G06F 9/45558 Hypervisor-specific management and integration aspects
• CPC Class Codes: G06F 9/5077 Logical partitioning of resources; Management or configuration of virtualized resources specific details on emulation or internal functioning of virtual machines G06F9/455
• CPC Class Codes: H04L 67/101 based on network conditions
• CPC Class Codes: H04L 67/131 Protocols for games, networked simulations or virtual reality

Abstract

Methods of offering network performance guarantees in multi-tenant datacenters are described. In an embodiment, a request for resources received at a datacenter from a tenant comprises a number of virtual machines and a performance requirement, such as a bandwidth requirement, specified by the tenant. A network manager within the datacenter maps the request onto the datacenter topology and allocates virtual machines within the datacenter based on the available slots for virtual machines within the topology and such that the performance requirement is satisfied. Following allocation, stored residual capacity values for elements within the topology are updated according to the new allocation and this updated stored data is used in mapping subsequent requests onto the datacenter. The allocated virtual machines form part of a virtual network within the datacenter which is allocated in response to the request and two virtual network abstractions are described: virtual clusters and virtual oversubscribed clusters.

Claims

1. A method of resource allocation within a datacenter comprising:

receiving a request for resources from a user, the request comprising a number of virtual machines and a user-specified performance characteristic;

mapping the request to physical resources within the datacenter to allocate the requested number of virtual machines on physical machines within the datacenter; and

if the allocation is successful, updating stored data identifying residual capacity on a set of elements within the datacenter as a result of the allocation of virtual machines to the user; and

sending a message to the user accepting the request.

2. A method according to claim 1, wherein the performance characteristic comprises a bandwidth characteristic on links between virtual machines and wherein the stored data identifying residual capacity comprises stored data identifying residual bandwidth on each link in the datacenter.

3. A method according to claim 1, wherein mapping the request to physical resources within the datacenter to allocate the requested number of virtual machines comprises mapping the request to allocate a virtual cluster to the user, the virtual cluster comprising the requested number of virtual machines and a virtual switch connecting all of the virtual machines in a one-level tree topology.

4. A method according to claim 1, wherein mapping the request to physical resources within the datacenter to allocate the requested number of virtual machines comprises:

starting at a first level in a topology tree within the datacenter, determining how many virtual machines can be allocated in each sub-tree in the level;

if a sub-tree in the level can accommodate all the requested virtual machines, allocating the requested number of virtual machines in the sub-tree to the user; and

if none of the sub-trees in the level can accommodate all the requested virtual machines, moving to a next level in the topology tree and repeating the determining step.

5. A method according to claim 4, wherein the performance characteristic comprises a bandwidth characteristic, B, on links between virtual machines and the set M_v of virtual machines m that can be allocated in each sub-tree, v, in a level is defined by:

M_v={mε

[0,min(k_v,N)]s·

t·

min(m,N−

m)*B<

R_l}

where N is the requested number of virtual machines, k_v is the number of empty slots in the sub-tree and R_l is the residual bandwidth for a link l.

6. A method according to claim 4, wherein if the request received from the user further comprises an oversubscription factor and a group size, the requested number of virtual machines are divided into groups of virtual machines of a size specified by the group size and the mapping step is performed for each group of virtual machines in turn.

7. A method according to claim 1, wherein the request further comprises an oversubscription factor and a group size, wherein mapping the request to physical resources within the datacenter to allocate the requested number of virtual machines comprises mapping the request to allocate a virtual oversubscribed cluster to the user, the virtual oversubscribed cluster comprising the requested number of virtual machines arranged in groups of the specified size, each group comprising virtual machines connected to a virtual group switch and wherein the virtual oversubscribed cluster further comprises a virtual root switch connected to all of the virtual group switches.

8. A method according to claim 7, wherein mapping the request to physical resources within the datacenter to allocate the requested number of virtual machines comprises, for each group:

starting at a first level in a topology tree within the datacenter, determining how many virtual machines can be allocated in each sub-tree in the level;

if a sub-tree in the level can accommodate all the virtual machines in the group, allocating those virtual machines in the sub-tree to the user; and

if none of the sub-trees in the level can accommodate all the virtual machines in the group, moving to a next level in the topology tree and repeating the determining step.

9. A method according to claim 8, wherein the performance characteristic comprises a bandwidth characteristic, B, on links between virtual machines and the set M_v,i of virtual machines that can be allocated to each sub-tree, v, for group i in a level is defined by:

M_v,i={g_iε

[0,min(k_v,S)]s·

t·

Σ

_j=1ⁱCB_r,j,l(i)≦

R_l}

where

CB_r,j,l(i)=min(g_j*B,(S−

g_j)*B+min(S*B/O,E))

E=Σ

<sub>k=l,k≠

j</sub>ⁱ(S−

g_k)*B+Σ

_k=i+1^PS*B

and g_i is the number of members of group i in the sub-tree, S is the requested number of virtual machines in a group, O is the oversubscription factor, k_v is the number of empty slots in the sub-tree and R_l is the residual bandwidth for a link l.

10. A system comprising a datacenter, the datacenter comprising a network manager and a plurality of machines connected by switches, each machine comprising at least one slot for allocation of a virtual machine to a tenant and wherein the network manager is arranged to receive a request from a tenant comprising a number of virtual machines requested and a tenant-specified performance characteristic, to allocate virtual machines to the tenant to satisfy the performance characteristic based on stored residual capacity data for the datacenter and to update the stored residual capacity data as a result of the new allocation.

11. A system according to claim 10, wherein the network manager comprises a data store arranged to store the residual capacity data, network topology information for the datacenter and a record of available virtual machine slots within the datacenter.

12. A system according to claim 11, wherein the data store is further arranged to store allocation information corresponding to previously allocated tenant requests and wherein the network manager is arranged to use the allocation information to reallocate virtual machines upon failure or network reconfiguration.

13. A system according to claim 10, wherein, in use, the datacenter further comprises a plurality of virtual networks, each virtual network corresponding to an allocated request from a tenant and comprising a plurality of virtual machines interconnected by at least one virtual switch.

14. A system according to claim 13, wherein at least one of the plurality of virtual networks comprises a number of virtual machines specified in the request and a virtual switch connecting to each of the virtual machines in a one-level tree topology

15. A system according to claim 13, wherein at least one of the plurality of virtual networks comprises a number of virtual machines specified in the request arranged in groups, each group comprising a number of virtual machines as specified in the request and each virtual machine in a group is connected to a virtual group switch and wherein the virtual group switches are all connected to a virtual root switch.

16. A system according to claim 10, wherein the network manager is arranged to allocate virtual machines to the tenant to satisfy the performance characteristic by:

starting at a first level in a topology tree within the datacenter, determining a number of virtual machines that can be allocated to each sub-tree in the level subject to the tenant-specified performance characteristic;

if a sub-tree in the level can accommodate all the requested number of virtual machines, allocating the requested number of virtual machines in the sub-tree to the tenant; and

if none of the sub-trees in the level can accommodate all the requested number of virtual machines, moving to a next level in the topology tree and repeating the determining step until a sub-tree is identified that can accommodate all the requested number of virtual machines.

17. A system according to claim 10, further comprising an enforcement module at each virtual machine, wherein the enforcement module is arranged to limit traffic to and from the virtual machine in accordance with the tenant-specified performance characteristic.

18. A system according to claim 17, wherein the performance characteristic comprises a bandwidth characteristic and wherein an enforcement module at a virtual machine is arranged to measure traffic rates from the virtual machine to other virtual machines and to perform per-destination-virtual-machine rate limiting to enforce calculated rate limits.

19. A system according to claim 10, further comprising a tenant device arranged to transmit the request to the datacenter.

20. A method of resource allocation within a datacenter comprising:

receiving a request from a user, the request comprising a requested number of virtual machines and a user-specified bandwidth characteristic;

determining if the request further comprises an oversubscription factor and a group size;

if the request does not comprise an oversubscription factor and a group size, mapping the request to physical resources within the datacenter to allocate the requested number of virtual machines on physical machines within the datacenter by;

starting at a first level in a topology tree within the datacenter, determining a number of virtual machines that can be allocated in each sub-tree in the level;

if a sub-tree in the level can accommodate all the requested number of virtual machines, allocating the requested number of virtual machines in the sub-tree to the user; and

if none of the sub-trees in the level can accommodate all the requested number of virtual machines, moving to a next level in the topology tree and repeating the determining step;

if the request does comprise an oversubscription factor and a group size, dividing the requested number of virtual machines into a plurality of groups according to the specified group size and performing the mapping step for each group of virtual machines in turn;

updating stored data identifying residual bandwidth on each link within the datacenter as a result of the allocation of the requested number virtual machines to the user; and

sending a message to the user accepting the request.