Method and system for dynamically allocating servers to compute-resources using capacity thresholds

• US 20070233866A1
• Filed: 03/28/2006
• Published: 10/04/2007
• Est. Priority Date: 03/28/2006
• Status: Abandoned Application

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Appleby, Karen, Goldszmidt, German
• Application Number:
  US11/390,369
• Publication Number:
  US 20070233866A1
• Time in Patent Office: Days
• US Class Current: 709/226
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers
• CPC Class Codes: H04L 67/1008 based on parameters of servers, e.g. available memory or workload monitoring of computer activity G06F11/30
• CPC Class Codes: H04L 67/101 based on network conditions
• CPC Class Codes: H04L 67/1012 based on compliance of requirements or conditions with available server resources
• CPC Class Codes: H04L 67/1023 based on a hash applied to IP addresses or costs

Abstract

Servers are allocated for use in one of a plurality of compute-resources or for stand-by storage in a free-pool. Server load metrics are selected (e.g., ping-reply time or CP utilization) for measuring load in the servers. Metrics are measured for the servers allocated to the compute-resources. Several metrics can be measured simultaneously. The metrics for each compute-resource are normalized and averaged. Then, the metrics for each compute-resource are combined using weighting coefficients, producing a global load value, G, for each compute-resource. The G value is recalculated at timed intervals. Upper and lower thresholds are set for each compute-resource, and the G values are compared to the thresholds. If the G value exceeds the upper threshold, then a server in the free-pool is reallocated to the compute-resource; if the G value is less than the lower threshold, then a server is moved from the compute-resource to the free-pool.

Claims

1. A load driven method for allocating servers among a plurality of compute-resources and a free-pool, wherein each compute-resource comprises a plurality of servers, the method comprising the steps of:

for each monitored metric on the standard server and for each compute-resource, calculating a maximum metric value at a maximum load point as a maximum load vector for a compute-resource;

setting lower and upper global thresholds as allowable percents of the maximum load point;

for each compute-resource and unique server type and for each monitored metric, calculating a capacity weight for the monitored metric;

monitoring each server allocated to a compute-resource for at least one metric;

for each monitored metric and for each compute-resource, calculating an average normalized metric value P_n in standard server units;

for each monitored metric and for each compute-resource, calculating a current percent of a corresponding maximum metric value as a current percent of maximum load vector;

for each compute-resource, calculating one or more global load values G, wherein each global load value is a linear combination of normalized current percent of corresponding maximum metric values;

for each compute-resource, dynamically adjusting lower upper thresholds for the global load value; and

for each compute-resource, comparing the calculated global load value G to the lower threshold and upper threshold, and performing an allocation of servers to compute-resources based on a comparison outcome.

2. The method of claim 1, wherein following the comparison outcome, if a load is not predicted to continue for more than some minimum amount of time, do nothing.

3. The method of claim 1, wherein following the comparison outcome, if some predetermined amount of time has not elapsed since a last capacity adjustment, do nothing.

4. The method of claim 1, wherein following the comparison outcome, if servers are available in the free pool and an overloaded compute-resource has a global load value G greater than the upper threshold, then removing a server from the free pool and allocating it to the overloaded compute-resource.

5. The method of claim 1, wherein following the comparison outcome, if servers are not available in the free pool and an overloaded compute-resource has a global value G greater than the upper threshold, perform resource-negotiation.

6. The method of claim 1, wherein following the comparison outcome, if an under loaded compute-resource has a global load value G less than the lower threshold, and the following inequality is satisfied

${\mathrm{Curr\_Total}}_{\mathrm{req}/\sec }<$

Server_Max req / sec * ( N - 1 ) N - ( Server_Max req / sec - Server_Min req / sec )

then removing a server from the under loaded compute-resource and allocating it to the free-pool.

7. The method of claim 1 wherein the maximum load values contained in the maximum-load-vector correspond to the values measured on the standard server when load reaches the response time transition point

Max LV_compute—

_resource=(M1_stdr, . . . , MN_stdr)

8. The method of claim 1, wherein a capacity weight of an nth metric on a given compute-resource is calculated according to the equation

${\mathrm{MWn}}_t=\frac{{\mathrm{Mn}}_t}{{\mathrm{Mn}}_{\mathrm{stdr}}}$

9. The method of claim 1, wherein each normalized average metric value P is calculated according to the equation

$P_n=(\frac{\sum _{}}{}$

s ∈

S ⁢

measured_valueM n ⁡

( s ) ∑

s ∈

S ⁢

MWn t ⁡

( s ) )

wherein P_n is the present value of metric n on server s in standard server units, m is the number of servers assigned to the compute resource.

10. The method of claim 1, wherein the Current Percent of Maximum Load Vector (% CurrMLV), is calculated according to the equation

$\%$

 

⁢

CurrMLV = ( P 1 M 1 ⁢

_stdr , …

⁢

 

, P n M n_stdr )

11. The method of claim 1, wherein one or more global load values G are computed for each compute-resource, as a linear combination of normalized current percent of the corresponding maximum values according to the following equation

$\mathrm{Load}=\sum _{}^{}$

n = 1 N ⁢

( C n * % ⁢

 

⁢

CurrMLV n )

12. The method of claim 1, wherein dynamic upper and lower thresholds for the global load value are adjusted using the following equation

$\mathrm{Threshold\_Adjustment}=\frac{\mathrm{Threshold\_High}-\mathrm{Threshold\_Low}}{\mathrm{Size\_Growth}}$

_Interval

13. The method of claim 1, wherein a deallocation process is inhibited unless following inequality is satisfied

${\mathrm{Curr\_Total}}_{\mathrm{req}}$

/ ⁢

sec <

Server_Max req ⁢

/ ⁢

sec * ( N - 1 ) N - ( Server_Max req ⁢

/ ⁢

sec - Server_Min req ⁢

/ ⁢

sec )

14. A computer readable medium containing code which enables a computer to perform a method for allocating servers among a plurality of connected compute-resources and a free-pool, wherein each compute-resource comprises a plurality of servers, the method comprising the steps of:

for each monitored metric on the standard server and for each compute-resource, calculating a maximum metric value at a maximum load point as a maximum load vector for the compute-resource;

monitoring each server allocated to a compute-resource for at least one metric;

for each monitored metric and for each compute-resource, calculating an average normalized metric value P_n in standard server units;

for each monitored metric and for each compute-resource, calculating a current percent of a corresponding maximum metric value as a current percent of maximum load vector;

for each compute-resource, calculating one or more global load values G, wherein each global load value is a linear combination of normalized current percent of the corresponding maximum metric values;

for each compute-resource, defining dynamically calculated lower threshold and an upper threshold adjustments for the global load value; and

for each compute-resource, comparing the calculated global load value G to the lower threshold and upper threshold, and performing a server allocation according to a comparison outcome.

15. The computer readable medium of claim 14, wherein the method, following the comparison outcome, determines if load is not predicted to continue for more then some minimum amount of time, and if so, does nothing.

16. The computer readable medium of claim 14, wherein the method, following the comparison outcome, determines if some predetermined amount of time has not elapsed since the last capacity adjustment, and if so, does nothing.

17. The computer readable medium of claim 14, wherein the method, following the comparison outcome, determines if servers are available in the free pool and an overloaded compute-resource has a global load value G greater than the upper threshold, and if so, removes a server from the free-pool and allocating it to the overloaded compute-resource.

18. The computer readable medium of claim 14, wherein the method, following the comparison outcome, determines if servers are not available in the free pool and an overloaded compute-resource has a global load value G greater than the upper threshold, and if so, performs resource-negotiation.

19. The computer readable medium of claim 14, wherein the method,

following the comparison outcome, determines if an under loaded compute-resource has a global load value G less than the lower threshold, and if so, removes a server from the under loaded compute-resource and allocating it to the free-pool.

20. A system for allocating servers among a plurality of connected server compute-resources and a free-pool, wherein each server compute-resource comprises a plurality of servers, the system comprising:

monitoring means for monitoring each server allocated to a compute-resource for a plurality of metric values;

calculating means for calculating a normalized average metric value P for each monitored metric value and for each server compute-resource;

combining means for linearly combining the normalized metric values to create a global load value G for each compute-resource;

storage means for storing a defined lower threshold and a defined upper threshold for the linear combination value;

comparing means for comparing the global load value to the lower threshold and upper threshold; and

allocating means for allocating servers among compute-resources and the free-pool.

21. The system of claim 20, wherein the allocating means responds to the comparing means in the case where an overloaded compute-resource has a global load value greater than the upper threshold by removing a server from the free-pool and allocating it to the overloaded compute-resource.

22. The system of claim 20, wherein the allocating means responds to the comparing means in the case where an under loaded compute-resource has a global load value less than the lower threshold by removing a server from the under loaded compute-resource and allocating it to the free-pool.

23. The system of claim 20, wherein the allocating means responds to the comparing means in the case where an under loaded compute-resource has a global load value G less than the lower threshold and an overloaded compute-resource has a global load value G greater than the upper threshold by removing a server from the under loaded compute-resource and allocating it to the overloaded compute-resource.

24. The system of claim 20, further comprising means for calculating a capacity weight of each server type for each compute-resource.

25. The system of claim 24, wherein server capacity weights are klused in combination with current metric values to compute a present load as represented by each metric type.

26. The system of claim 20, wherein a Current Percent Maximum Load vector is linearly combined with metric reliability weights to generate one or more global compute-resource weights for each compute-resource.

27. The system of claim 20, wherein each compute-resource upper and lower thresholds are dynamically adjusted.