Senior / Lead Performance Engineer | Load testing | Capacity planning and NFR
🌐 English: B2 — professional working proficiency
📄 Resume: CV (full) 🔗 LinkedIn: https://www.linkedin.com/in/georgii-kaliaev-3826b6214/
📨 Telegram: @KGeorgeK
📦 Open source: jmeter-load-profile-checker
I am a Senior / Lead Performance Engineer with 7+ years of hands-on experience building and running load, stress, capacity, and scalability testing for high-load distributed systems — fintech, banking, e-commerce, insurance, government, and trading.
I design workload models and NFRs (p95/p99, throughput), conduct test campaigns, build observability around them, find bottlenecks, and deliver actionable recommendations to development and architecture.
At IBS I led a load-testing team of ~5 engineers (Team Player 2023, Project Driver 2022) — e.g. identified and closed ~500+ SAP defects before prod (~4 000 users), agreed Treasury LTM ~45k ops/h, ran grocery night LT +189% with 0 incidents, validated exchange ~10k WebSocket msg/s, validated Guidewire ~165k ops/h.
Currently at First Asset Management (similar to Vanguard) as Principal Development Engineer (load testing): introduced Kangal + JMeter on Kubernetes, load-tested 15+ REST microservices (on-demand injectors, scaling, teardown — no idle load hardware), built PostgreSQL capacity models (100M+ rows), reduced total CPU by 100+ vCPU, and developed internal Python tooling in GitLab CI.
- Cut total CPU usage by 100+ vCPU through performance tuning.
- Built Kangal + JMeter on Kubernetes and load-tested 15+ microservices with on-demand generators.
- Developed jmeter-load-profile-checker — campaign analysis ~5–6 h → ~1 h.
- Conducted grocery prod night LT: ~900 → 2600+ orders/h (+189%), 0 incidents.
- Validated exchange load at ~10,000 WebSocket msg/s and found a thread leak (OOM risk removed pre go-live).
- Identified and closed ~500+ SAP performance defects before production.
- Led load-testing team of ~5 at IBS.
Outcomes from load campaigns — most significant cases only. Full detail: resume.md.
Asset management firm (funds / portfolios) — internal product platform (Sber Group-related projects), not client consulting.
| Area | Outcome |
|---|---|
| PostgreSQL / capacity | Built 5 growth models (0→100M+ rows). Forecast SLA risk (~90% accuracy). Identified INSERT degradation up to 8.3× (~2.7 s → 22–77 s). Proposed partitioning — after fix INSERT 20–70 s → ≤1 s (to 2B rows); horizon ~11 months → ~30 years — write-up |
| K8s / Kangal | Introduced Kangal + JMeter as the default LT platform. Load-tested 15+ REST services with on-demand injectors, worker scaling, teardown. No idle LT hardware between campaigns |
| CPU / tuning | Reduced total CPU by 100+ vCPU through performance tuning from LT findings |
| Tooling | Developed and rolled out jmeter-load-profile-checker — analysis ~5–6 h → ~1 h per campaign |
| Batch / scaling | Showed throughput capped by 1 min cron, not pod count (1→3 pods — no gain). Recommended event-driven / worker pool |
| Client / project | Outcome |
|---|---|
| Cooper / SberMarket grocery e-commerce / retail delivery |
Developed Gatling scenarios (B2C/B2B + mobile) and ran prod night LT (23:00–03:00) with 0 incidents: ~900 → 2600+/h (+189%). 400% profile ~3900 HTTP req/s, 5000+ users, 30+ scenarios — suite |
| SPIMEX commodity exchange, St. Petersburg |
Conducted trading load (WebSocket/STOMP + REST) at ~10 000 msg/s. Found soak thread leak (VisualVM) — OOM risk removed pre go-live |
| Federal Treasury / GIIS gov. public finance / e-budget |
Agreed LTM ~45k ops/h. Built GOST / CryptoPro contour and implemented NGINX (GOST certs per port, Lua logging) — article |
| Rosgosstrakh / Guidewire insurance (PolicyCenter) |
Developed end-to-end scenarios and 15+ stubs. Validated peak ~165k ops/h at 80% profile before launch |
| Leroy Merlin TMS DIY retail - transport / carrier portal |
Developed Gatling tests and signed off 1200% load. Caught PG query 32 ms → 15.1 min on ~233k shipments volume before go-live — suite |
| Sberbank SAP largest Russian bank - ERP / BW |
Ran LoadRunner / Performance Center for ~4 000 users and ~110 scripts. Identified and closed ~500+ performance defects before go-live |
| Megapolis / IBS portal SAP HR portal + corporate Bitrix |
Conducted capacity sign-off. Bitrix auth ~425 → ~850 logins/15 min. Found threads not shutting down under soak |
| JVM / VisualVM cross-engagement finding |
Monitored soak with remote VisualVM; found threads not shutting down; fix applied right after LT findings |
At that time ScriptMaster delivered Alfa-Bank projects only. Two engagements:
| Project | Outcome |
|---|---|
| FSSP (Federal Bailiff Service) | Built LT from scratch. Conducted LoadRunner campaigns up to ~5000 VU (IBM MQ + file-drop XML over SMB/FTP). Developed ~30 SOAP stubs. Fixed HornetQ backlog. Improved Oracle latency ~−25%, throughput ~+15% |
| Citrix / RDP | Conducted remote-desktop load tests and evaluated channel/server stability under peak concurrent sessions |
- Design realistic load models and NFR / methodology from business traffic and production signals
- Build performance frameworks (JMeter, Gatling, LoadRunner, k6); run Kangal on Kubernetes (on-demand injectors, scaling, teardown); tune load-generator hosts (
limits.conf/sysctl) - Optimize and extend JMX / VuGen / Gatling scenarios with Groovy, Bash, Redis and REST API — shorter prep, more realistic synthetic flow
- Conduct load, stress, soak, and capacity tests; integrate regression performance into release cycles
- Correlate Grafana / Prometheus / ELK / APM with test windows; perform RCA across app, JVM, DB, Kafka, K8s
- Perform capacity planning and SLA forecasting (PostgreSQL at scale); drive tuning that cut 100+ vCPU on the current platform
- Lead and mentor performance engineers (team of ~5 at IBS); conduct technical interviews and hiring
- Deliver clear recommendations to engineering, DevOps, and stakeholders
Load testing
Observability / APM
Databases and cache
Messaging
Protocols
Languages
Cloud / DevOps
Middleware
Stubs / integration
JVM and OS analysis
Tooling
- Banking and FinTech
- Insurance
- E-commerce and Logistics
- Commodity Trading Platforms
- Government Financial Systems
- Conduct load, stress, and capacity testing for cloud-native / microservices systems
- Validate NFRs (p95/p99, throughput, error rate, scalability KPIs)
- Perform bottleneck analysis (application, JVM, database, Kafka/RabbitMQ, K8s, LB)
- Build performance monitoring / observability and correlate distributed traces with test windows
- Integrate performance pipelines into CI/CD (GitLab CI, Jenkins, Bamboo)
- Lead performance engineers (team of ~5 at IBS): mentoring, interviews, campaign standards
- Improve stability, scalability, and reliability of high-load platforms
- jmeter-load-profile-checker (open source) — validate JMeter load profile by step; campaign analysis ~5–6 h → ~1 h
- PostgreSQL bottleneck at scale — capacity, INSERT degradation, data model and indexing — LinkedIn
- NGINX + JMeter — multiple client certs, GOST, proxy instead of Fiddler — LinkedIn
- Load generator OS tuning —
limits.conf/sysctlwhen the bottleneck is the injector, not the SUT — LinkedIn
I am interested in international projects where performance, scalability, and system reliability are critical.
Open to roles focused on Performance Engineering, Non-Functional Testing, or System Optimization.
Users
500 ┤ ╭──────────╮
400 ┤ ╭────╯ │
300 ┤ ╭────╯ │
200 ┤ ╭────╯ │
100 ┼──────────╯ ╰──────────
0s 5m 10m 15m 20m
Response Time increases after saturation point
“Performance problems are rarely about one slow component. They are about how systems behave under pressure.”