Predicting Performance Anomalies in Software Systems at Run-time

2021 ◽  
Vol 30 (3) ◽  
pp. 1-33
Author(s):  
Guoliang Zhao ◽  
Safwat Hassan ◽  
Ying Zou ◽  
Derek Truong ◽  
Toby Corbin

High performance is a critical factor to achieve and maintain the success of a software system. Performance anomalies represent the performance degradation issues (e.g., slowing down in system response times) of software systems at run-time. Performance anomalies can cause a dramatically negative impact on users’ satisfaction. Prior studies propose different approaches to detect anomalies by analyzing execution logs and resource utilization metrics after the anomalies have happened. However, the prior detection approaches cannot predict the anomalies ahead of time; such limitation causes an inevitable delay in taking corrective actions to prevent performance anomalies from happening. We propose an approach that can predict performance anomalies in software systems and raise anomaly warnings in advance. Our approach uses a Long-Short Term Memory neural network to capture the normal behaviors of a software system. Then, our approach predicts performance anomalies by identifying the early deviations from the captured normal system behaviors. We conduct extensive experiments to evaluate our approach using two real-world software systems (i.e., Elasticsearch and Hadoop). We compare the performance of our approach with two baselines. The first baseline is one state-to-the-art baseline called Unsupervised Behavior Learning. The second baseline predicts performance anomalies by checking if the resource utilization exceeds pre-defined thresholds. Our results show that our approach can predict various performance anomalies with high precision (i.e., 97–100%) and recall (i.e., 80–100%), while the baselines achieve 25–97% precision and 93–100% recall. For a range of performance anomalies, our approach can achieve sufficient lead times that vary from 20 to 1,403 s (i.e., 23.4 min). We also demonstrate the ability of our approach to predict the performance anomalies that are caused by real-world performance bugs. For predicting performance anomalies that are caused by real-world performance bugs, our approach achieves 95–100% precision and 87–100% recall, while the baselines achieve 49–83% precision and 100% recall. The obtained results show that our approach outperforms the existing anomaly prediction approaches and is able to predict performance anomalies in real-world systems.

Author(s):  
Jürgen Dorn ◽  
Stefan Labitzke

We present an analytical approach to detect relationships between persons in the real world, such as friendship, rivalry, or others, out of the behavior of members in a social software system. In social software systems, users often evaluate submissions of other users. If these actions are somehow biased, we assume a personal relationship between these users. If we know about the relationship between two users, the validity of the evaluation, and with that, the trust into the social software, can be improved. For example, if a rival evaluates a submission unfairly, we should decrease the impact of this evaluation. We apply the approach in TechScreen, a social software system that supports the exchange of knowledge about Internet technologies. Since we try to mine competencies of its users, the validity of evaluations is very important. In this chapter we show results of experiments with about 50 users.


Author(s):  
Jürgen Dorn ◽  
Stefan Labitzke

We present an analytical approach to detect relationships between persons in the real world, such as friendship, rivalry, or others, out of the behavior of members in a social software system. In social software systems, users often evaluate submissions of other users. If these actions are somehow biased, we assume a personal relationship between these users. If we know about the relationship between two users, the validity of the evaluation, and with that, the trust into the social software, can be improved. For example, if a rival evaluates a submission unfairly, we should decrease the impact of this evaluation. We apply the approach in TechScreen, a social software system that supports the exchange of knowledge about Internet technologies. Since we try to mine competencies of its users, the validity of evaluations is very important. In this chapter we show results of experiments with about 50 users.


Blood ◽  
2020 ◽  
Vol 136 (Supplement 1) ◽  
pp. 17-18
Author(s):  
Parameswaran Hari ◽  
Lita Araujo ◽  
Dominick Latremouille-Viau ◽  
Peggy Lin ◽  
Mikhail Davidson ◽  
...  

Background: Renal impairment (RI) is associated with substantial clinical and economic burden in patients with multiple myeloma (MM), but real-world data reporting on healthcare resource utilization (HRU) and outcomes in these patients are lacking. We assessed treatment patterns, overall survival (OS), HRU and associated costs across lines of therapy (LoT) in patients with MM who had baseline RI. Methods: We identified patients (aged ≥18 years) with continuous Part A, B and D coverage who initiated pharmacologic therapy for MM between January 1, 2012 and December 31, 2016. Baseline demographics, disease characteristics, and treatment patterns from first-line to fourth-line (1L-4L) were reported for all eligible patients (main cohort). Within this cohort, a subgroup of patients diagnosed with RI at baseline (RI subgroup) were identified using appropriate International Classification of Diseases (ICD)-9 and ICD-10 codes. Treatment regimens were identified during the first 60 days following start of each LoT; stem cell transplantation (SCT) in 1L was considered part of the 1L regimen. The end of each LoT was indicated by treatment augmentation, treatment switching (after >60 days), discontinuation of all agents (for >90 days), or death. Overall survival (Kaplan-Meier analysis) was defined as time from start of each LoT until death or censoring (end of data/Medicare coverage). All-cause HRU categories were identified during each LoT and reported as incidence rate per patient per month (PPPM); associated all-cause healthcare costs during LoT were reported in 2017 US$. Results are presented using standard descriptive statistics. Results: A main cohort of 10,026 patients was identified; of these, a RI subgroup of 714 patients with baseline RI was identified (7.1% of main cohort). At 1L initiation, the RI subgroup was generally younger (71.9 vs. 74.6 years), had a lower proportion of females (47.8% vs. 53.1%) and had a higher proportion of Medicare coverage for end-stage renal disease (62.9% vs. 6.3%) than the main cohort. Patients with RI had a higher mean Charlson Comorbidity Index score (excluding MM; 4.8 vs. 3.3) and a higher proportion of patients with comorbidities (anemia: 72.5% vs. 57.9%; diabetes with chronic complications: 38.7% vs. 27.1%; cardiovascular diseases: 97.2% vs. 82.5%) than the main cohort. In the RI subgroup, among patients who received SCT in 1L (n=76), bortezomib-dexamethasone (Vd) was the most frequent 1L regimen (39.5%), followed by bortezomib-lenalidomide-dexamethasone (VRd; 17.1%) and bortezomib-cyclophosphamide-dexamethasone (VCd; 15.8%). In patients who had no SCT in 1L, Vd was the most frequent 1L regimen (59.5%), followed by VCd (12.7%) and lenalidomide-dexamethasone (Rd; 12.1%). Among patients in the RI subgroup who progressed to 2L therapy, 61.7% received lenalidomide-based regimens in 1L. Newer MM therapies such as carfilzomib, pomalidomide, ixazomib, daratumumab, and elotuzumab were used more frequently in later LoTs (2L: 25.6%; 3L: 50.0%; 4L: 68.8%). Median OS from start of 1L was shorter in the RI subgroup than in the main cohort (29.9 vs. 46.5 months; Table), and this difference was consistent across each subsequent LoT. Incidence of HRU during 1L (Table) was generally higher in the RI subgroup than the main cohort, particularly for inpatient days (1.3 vs. 0.7 PPPM) and home health services (0.9 vs. 0.5 PPPM); this pattern was consistent between cohorts across each subsequent LoT. Total costs in the 1L RI subgroup vs. main cohort (Table) were $14,782 vs. $12,451; the cost differential was maintained across each subsequent LoT. The key driver of this difference was the additional medical service costs ($12,047 vs. $7,459 in 1L) incurred by patients with RI. Conclusion: Patients with MM who had baseline RI were shown to experience higher clinical and economic burden in real-world clinical practice than the overall MM population. This burden was maintained across LoTs. Efficacious regimens that help improve renal function with minimal toxicity would enable patients with MM and RI to persist with treatment and may help address unmet need in this subgroup of patients. Table Disclosures Hari: BMS: Consultancy; GSK: Consultancy; Janssen: Consultancy; Amgen: Consultancy; Takeda: Consultancy; Incyte Corporation: Consultancy. Araujo:Sanofi Genzyme: Current Employment. Latremouille-Viau:Sanofi Genzyme: Consultancy, Other: Dominique Latremouille-Viau is an employee of Analysis Group, Inc. which received consultancy fees from Sanofi Genzyme.; Novartis Pharmaceutical Corporation: Consultancy, Other: Dominique Latremouille-Viau is an employee of Analysis Group, Inc. which received consultancy fees from Novartis.. Lin:Sanofi Genzyme: Current Employment. Davidson:Sanofi Genzyme: Other: Mikhail Davidson is an employee of Analysis Group, Inc which received consultancy fees from Sanofi Genzyme.. Guerin:Sanofi Genzyme: Consultancy, Other: Annie Guerin is an employee of Analysis Group, Inc. which received consultancy fees from Sanofi Genzyme.; Abbvie: Consultancy, Other; Novartis Pharmaceuticals Corporation: Consultancy, Other: Annie Guerin is an employee of Analysis Group, Inc. which received consultancy fees from Novartis.. Sasane:Sanofi Genzyme: Current Employment.


2018 ◽  
Vol 9 (1) ◽  
pp. 204589401881629 ◽  
Author(s):  
Sean Studer ◽  
Michael Hull ◽  
Janis Pruett ◽  
Eleena Koep ◽  
Yuen Tsang ◽  
...  

Several new medications for pulmonary arterial hypertension (PAH) have recently been introduced; however, current real-world data regarding US patients with PAH are limited. We conducted a retrospective administrative claims study to examine PAH treatment patterns and summarize healthcare utilization and costs among patients with newly diagnosed PAH treated in US clinical practice. Patients newly treated for PAH from 1 January 2010 to 31 March 2015 were followed for ≥12 months. Patient characteristics, treatment patterns, healthcare resource utilization, and costs were described. Adherence (proportion of days covered), persistence (months until therapy discontinuation/modification), and the probability of continuing the index regimen were analyzed by index regimen cohort (monotherapy versus combination therapy). Of 1637 eligible patients, 93.8% initiated treatment with monotherapy and 6.2% with combination therapy. The most common index regimen was phosphodiesterase type 5 inhibitor (PDE-5I) monotherapy (70.0% of patients). A total of 581 patients (35.5%) modified their index regimen during the study. Most patients (55.4%) who began combination therapy did so on or within six months of the index date. Endothelin receptor agonists (ERAs) and combination therapies were associated with higher adherence than PDE-5Is and monotherapies, respectively. Healthcare utilization was substantial across the study population, with costs in the combination therapy cohort more than doubling from baseline to follow-up. The majority of patients were treated with monotherapies (most often, PDE-5Is), despite combination therapies and ERAs being associated with higher medication adherence. Index regimen adjustments occurred early and in a substantial proportion of patients, suggesting that inadequate clinical response to monotherapies may not be uncommon.


2018 ◽  
Vol Volume 10 ◽  
pp. 629-641 ◽  
Author(s):  
Machaon Bonafede ◽  
Joseph Feliciano ◽  
Qian Cai ◽  
Virginia Noxon ◽  
Nicole Princic ◽  
...  

Sign in / Sign up

Export Citation Format

Share Document