Phase-space surface representation of server computer performance in a computer network

• US 5,664,106 A
• Filed: 02/09/1995
• Issued: 09/02/1997
• Est. Priority Date: 06/04/1993
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Digital Equipment Corporation (HP Inc.)
• Inventors: Caccavale, Frank Samuel
• Primary Examiner(s): Heckler, Thomas M.
• Application Number:
  US08/385,606
• Time in Patent Office: 936 Days
• Field of Search: 395/200, 395/183.22, 395/180, 395/200.1, 395/200.11, 395/200.03, 395/183.13, 395/183.15, 395/184.01, 395/119, 364/550, 364/551.01
• US Class Current: 709/224
• CPC Class Codes: G06F 11/32 with visual or acoustical indication of the functioning of the machine
• CPC Class Codes: G06F 11/323 Visualisation of programs or trace data
• CPC Class Codes: G06F 11/328 Computer systems status display G06F11/327 takes precedence
• CPC Class Codes: G06F 11/3419 by assessing time
• CPC Class Codes: G06F 11/3433 for load management allocation of a server based on load conditions G06F9/505; load rebalancing G06F9/5083; redistributing the load in a network by a load balancer H04L67/1029
• CPC Class Codes: G06F 11/3452 Performance evaluation by statistical analysis
• CPC Class Codes: G06F 11/3495 for systems
• CPC Class Codes: G06F 12/0866 for peripheral storage systems, e.g. disk cache
• CPC Class Codes: G06F 2201/805 Real-time
• CPC Class Codes: G06F 2201/885 Monitoring specific for caches
• CPC Class Codes: G06F 2212/601 Reconfiguration of cache memory
• CPC Class Codes: H04L 43/00 Arrangements for monitoring or testing data switching networks
• CPC Class Codes: H04L 43/12 Network monitoring probes
• CPC Class Codes: H04L 43/16 Threshold monitoring
• CPC Class Codes: H04L 67/01 Protocols
• CPC Class Codes: H04L 9/40 Network security protocols
• CPC Class Codes: Y10S 707/99931 Database or file accessing

Abstract

In a method and system for dynamically improving the performance of a server in a network, a tuning system monitors a workload of the server in real time, monitors a set of internal performance characteristics of the server in real time, and monitors a set of adjustable server parameters of the server in real time. The workload of the server may include the frequency and type of service requests received by the server from clients in the network. The internal server performance characteristics may include, for example, a data cache hit ratio of a data cache in the server. The set of server parameters may include, for example, the overall data cache size or the data cache geometry of the server. The tuning system periodically alters one or more of the set of adjustable server parameters as a function of the workload and internal performance characteristics of the server. Since the tuning system is continuously monitoring workload and performance characteristics of the server and altering the server parameters accordingly, the effectiveness of a given change in the server parameters is reflected in the next set of monitored workload and performance values.

Claims

1. A method of operating a distributed computer system having client computers connected to a server computer, comprising the steps of:

tabulating the average arrival rates of primitive requests emanating from the client computers and directed to the server computer during sampling intervals of operation of the computer system, the primitive requests being of different types and their average arrival rates being tabulated by request type, the set of primitive requests emanating from the client computers during each sampling interval collectively constituting a workload for that interval, and the set of tabulated arrival rates for the different primitive request types during each sampling interval being arranged to make up a workload vector for that interval;

mapping the workload vectors to scalar workload values in a manner substantially guaranteeing that similar workloads map to similar workload values and that dissimilar workloads map to dissimilar workload values;

monitoring the average utilization by each workload of a resource in the server computer;

monitoring the average response time of the server computer to the requests in each workload;

forming for each workload a corresponding 3-dimensional phase-space vector (mapped scalar workload value, monitored average resource utilization, monitored average response time); and

plotting the formed 3-dimensional phase-space vectors in a corresponding 3-dimensional phase space.

2. A method according to claim 1, wherein the primitive request types comprise sequential reads, random reads, sequential writes, random writes, file opens, file closes, file creates, byte range locks, and directory searches.

3. A method according to claim 1, wherein the resource whose utilization is monitored is CPU time.

4. A method according to claim 1, wherein the mapping step comprises the steps of:

selecting a reference phase-space vector in the phase space;

computing the angle between the reference phase-space vector and each of the workload vectors, the computed angle for each workload vector representing the scalar workload value to which the workload vector is mapped;

identifying as collisions those sets of different workload vectors that are mapped to the same scalar workload value;

determining for each collision whether the commonly-mapped workload vectors represent similar workloads having similar sets of tabulated arrival rates of the primitive requests; and

if more than a predetermined small minority of the collisions are among dissimilar workloads, repeating the foregoing steps using different reference phase-space vectors until either (1) no more than the predetermined small minority of the collisions are among dissimilar workloads, or (2) the reference phase-space vectors have been exhausted, each repetition using a different phase-space vector resulting in a different mapping of workload vectors to scalar workload values;

and wherein the mapped scalar workload value used in the forming step is that yielded by the first-encountered mapping for which no more than the predetermined small minority of the collisions are among dissimilar workloads, if such a mapping exists, or a mapping for which the number of collisions among dissimilar workloads is least, if all mappings result in more than the predetermined small minority of collisions among dissimilar workloads.

5. A method according to claim 4, wherein the predetermined small minority is 5%.

6. A method according to claim 1, wherein the workloads are real-time workloads and the plotted phase-space vectors are real-time phase-space vectors collectively defining a real-time trajectory in the phase space during an operational phase of the server computer, and further comprising the step of plotting a set of characteristic phase-space vectors in the phase space, the set of characteristic phase-space vectors (representing mapped workload value, monitored average resource utilization, and monitored average response time) obtained when the server computer was previously operating in a characterization phase during which a set of characterizing workloads emanated from the client computers directed to the server computer.

7. A method according to claim 6, wherein the set of characteristic and real-time phase-space vectors collectively define a characteristic surface for the server computer in the phase space, and further comprising the steps of:

predicting, based on the real-time trajectory and on the geometry of the characteristic surface near the most recently plotted real-time phase-space vector, a future workload value representing a future workload that the server computer might be expected to service;

locating a point corresponding to the predicted future workload value on the characteristic surface; and

estimating a future response time and a future resource utilization that might be expected from the server computer based on the corresponding values for those operating characteristics at the located point.

8. A method according to claim 6, wherein the set of characteristic and real-time phase-space vectors collectively define a characteristic surface for the server computer in the phase space, and further comprising the steps of:

controlling, during each of successive validation intervals, the manner in which requests emanate from the client computers so that the set of resulting workload vectors in response to which the server computer operates substantially matches a predetermined set of validating workload vectors deemed useful for validating that the characteristic surface accurately represents the current relationships among workloads, average response times, and resource utilization values for the server computer; and

determining whether the phase-space vectors plotted in response to the validating workload vectors during one validation interval are significantly different from the same phase-space vectors plotted during a preceding validation interval.

9. A method according to claim 8, further comprising the step of notifying a user if the plotted phase-space vectors are determined to be significantly different in the two validation intervals.

10. A method according to claim 6, wherein the set of characteristic and real-time phase-space vectors collectively define a characteristic surface for the server computer in the phase space, and further comprising the steps of:

controlling, during each of successive validation intervals, the manner in which requests emanate from the client computers so that the set of resulting workload vectors in response to which the server computer operates substantially matches a predetermined set of validating workload vectors deemed useful for validating that the characteristic surface accurately represents the current relationships among workloads, average response times, and resource utilization values for the server computer;

determining whether the phase-space vectors plotted in response to the validating workload vectors during one validation interval are significantly different from the same phase-space vectors plotted during a preceding validation interval;

entering a re-characterization phase if the plotted phase-space vectors are determined to be significantly different in the two validation intervals;

controlling the manner in which requests emanate from the client computers during the re-characterization phase so that the set of resulting workload vectors in response to which the server computer operates substantially matches a predetermined set of workload vectors deemed useful for characterizing the operation of the server computer, the operation of the server computer during the re-characterization phase yielding a new set of characteristic phase-space vectors; and

plotting the new set of characteristic phase-space vectors in place of the previously-plotted characteristic phase-space vectors to define a new characteristic surface for the server computer.

11. A method according to claim 1, further comprising the step of controlling the manner in which requests emanate from the client computers so that the resulting set of workload vectors in response to which the server computer operates substantially matches a predetermined set of workload vectors.

12. A method according to claim 11, wherein the controlling step is performed during a characterization phase of operation of the settler computer and the predetermined set of workload vectors are characterizing workload vectors deemed useful for characterizing the operation of the server computer.

13. A method according to claim 12, further comprising the step of allowing requests to emanate from the client computers in an uncontrolled manner in response to user activity during an operational phase of the server computer succeeding the characterization phase, the resulting plotted phase-space vectors being real-time phase-space vectors collectively defining a real-time trajectory in the phase space, the set of characteristic and real-time phase-space vectors collectively defining a characteristic surface for the server computer in the phase space.

14. A method of operating a distributed computer system having client computers connected to a server computer comprising the steps of:

tabulating the average arrival rates of primitive requests emanating from the client computers and directed to the server computer during sampling intervals of operation of the computer system, the primitive requests being of different types and their average arrival rates being tabulated by request type, the set of primitive requests emanating from the client computers during each sampling interval collectively constituting a workload for that interval, and the set of tabulated arrival rates for the different primitive request types being arranged to make up a workload vector for that interval;

selecting a reference phase-space vector for a 3-dimensional phase space in which phase-space vectors are to be plotted;

computing the angle between the reference phase-space vector and each of the workload vectors, the computed angle for each workload vector representing a scalar workload value to which the workload vector is mapped;

identifying as collisions those sets of different workload vectors that are mapped to the same scalar workload value;

determining for each collision whether the commonly-mapped workload vectors represent similar workloads having similar sets of tabulated arrival rates of the primitive requests;

if more than a predetermined small minority of the collisions are among dissimilar workloads, repeating the selecting, computing, identifying and determining steps using different reference phase-space vectors until either (1) no more than the predetermined small minority of the collisions are among dissimilar workloads, or (2) the reference phase-space vectors have been exhausted, each repetition using a different phase-space vector resulting in a different mapping of workload vectors to scalar workload values;

monitoring the average utilization by each workload of a resource in the server computer;

monitoring the average response time of the server computer to the requests in each workload;

forming for each workload a corresponding 3-dimensional phase-space vector (mapped scalar workload value, monitored average resource utilization, monitored average response time), the mapped scalar workload value being that yielded by the first-encountered mapping for which no more than the predetermined small minority of the collisions are among dissimilar workloads, if such a mapping exists, or a mapping for which the number of collisions among dissimilar workloads is least, if all mappings result in more than the predetermined small minority of collisions among dissimilar workloads; and

plotting the formed 3-dimensional phase-space vectors in the phase space.

15. A method of operating a distributed computer system having client computers connected to a server computer, comprising the steps of:

controlling the manner in which primitive requests emanate from the client computers directed to the server computer during sampling intervals of operation of the computer system, the primitive requests being of different types, the set of primitive requests emanating from the client computers during each sampling interval collectively constituting a workload for that interval, the emanation of the primitive requests being controlled in a manner ensuring that the set of arrival rates at the server computer of each primitive request type during each sampling interval when arranged to make up a workload vector for that interval substantially matches a corresponding one of predetermined workload vectors;

mapping the workload vectors to scalar workload values in a manner substantially guaranteeing that similar workloads map to similar workload values and that dissimilar workloads map to dissimilar workload values;

monitoring the average utilization by each workload of a resource in the server computer;

monitoring the average response time of the server computer to the requests in each workload;

forming for each workload a corresponding 3-dimensional phase-space vector (mapped scalar workload value, monitored average resource utilization, monitored average response time); and

plotting the formed 3-dimensional phase-space vectors in a corresponding 3-dimensional phase space.

16. A method according to claim 15, wherein the controlling step is performed during a characterization phase of operation of the server computer and the predetermined set of workload vectors are characterizing workload vectors useful for characterizing the operation of the server computer.

17. A method according to claim 16, further comprising the steps of:

tabulating the average arrival rates of the primitive requests emanating from the client computers and directed to the server computer during sampling intervals of an operational phase of the server computer succeeding the characterization phase, the average arrival rates of the primitive requests being tabulated by request type, the set of primitive requests emanating from the client computers during each sampling interval collectively constituting a real-time workload for that interval, and the set of tabulated arrival rates for the different primitive request types during each sampling interval being arranged to make up a real-time workload vector for that interval; and

repeating the mapping, monitoring, forming and plotting steps of claim 15 for the real-time workloads and the real-time workload vectors.