Mohammed Mudasser

AI/ML Engineer

[Mohammed Mudasser @Md_mudasser]

As an Indian, it’s amazing to see @UPI_NPCI highlighted in tech! 🇮🇳

@georg_kreuzmayr connects India’s 10,000x growth in UPI transactions with the need for databases like @TigerBeetleDB, built for fast, reliable financial transactions.

@P99CONF @ScyllaDB
#P99CONF #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

What makes an object storage client fast?
@georg_kreuzmayr keynote:
🔹 Async I/O with io_uring
🔹 Batching boosts TCP receive throughput from 0.6 to 5.6 GB/s
🔹 Purpose-built, high-performance cloud I/O.
@P99CONF @ScyllaDB
#P99CONF #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

HPC performance monitoring, without the extra baggage. ⚡

Ershaad Basheer’s approach combines eBPF with LDMS to collect kernel-level metrics and track filesystem latency in real time.

Measure the system without making measurement the bottleneck.

@P99CONF @ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

100B vectors. 200ms p99 latency. 🤯
@natevanben at #P99CONF :
🔹 Hierarchical clustering speeds up search.
🔹 RaBitQ compresses vectors 16–32× while preserving recall.
🔹 Caching + sharding enable massive scale.
Serious performance engineering! 🚀
@P99CONF @ScyllaDB #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

Key takeaways from @curious_coding session at #P99CONF :

🔹 Memory access bottlenecks priority queues.
🔹 SIMD-friendly layouts beat traditional heaps.
🔹 SimdQuickHeap is up to 2× faster than radix heaps.

Data layout matters! ⚡

@P99CONF @ScyllaDB #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

Can AI improve cache algorithms?

@DorLaor keynote explored W-TinyLFU, combining LRU and SLRU to balance burst protection with long-term frequency.

Key takeaway: Smarter cache admission can improve hit rates and protect performance during scans.

@P99CONF @ScyllaDB
#P99CONF

[Mohammed Mudasser @Md_mudasser]

Me: “I’ll attend a few P99 CONF sessions and sleep early.” 😴
Also me at 11 PM: learning SIMD, 100B vector search, QUIC, and database optimizations… while tweeting takeaways. 😂
Performance matters. So does sleep. Apparently, I chose performance. 💀
@P99CONF @ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

Great insights from Felipe Mendes and Guilherme Nogueira at P99 CONF 2026!

📊 Live dashboards bring visibility into:
🔹 Request throughput
🔹 Read/write activity
🔹 Tablet distribution and load
@P99CONF @ScyllaDB #ScyllaDB #P99CONF #GRAFANA

[Mohammed Mudasser @Md_mudasser]

TSDB takeaways from @hagen1778 :

🔹 Shard buffers to reduce write contention.
🔹 Merge parts for better compression.
🔹 Filter by series ID and timestamp.
🔹 Process blocks concurrently for faster reads.
@P99CONF @ScyllaDB
#ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

The real challenge in TSDB performance? Making every stage of the data pipeline count.

@hagen1778 's approach:
⚡ Buffer writes in memory
⚡ Persist and merge data efficiently
⚡ Parallelize CPU-heavy query processing
Great insights into database engineering!
@P99CONF @ScyllaDB

[Mohammed Mudasser @Md_mudasser]

Building a fast time-series database is about more than ingesting data. It’s about how efficiently you store, filter and process it.

@hagen1778 highlights smart data layouts, efficient reads and performance optimization.

@P99CONF @ScyllaDB #ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

There’s no universal winner in analytical query performance!
Daniel Seybold’s benchmark insights:
🔹 Compares PostgreSQL, pg_duckdb, Citus and ClickHouse.
🔹 Evaluates ClickBench and TPC-DS.
🔹 Best performance depends on workload.
@P99CONF @ScyllaDB #P99CONF @duckdb @PostgreSQL

[Mohammed Mudasser @Md_mudasser]

Can PostgreSQL handle analytics with minimal code changes?
Daniel Seybold highlights:
🔹 pg_duckdb brings DuckDB’s vectorized execution to PostgreSQL.
🔹 Supports analytics and data lake integration.
🔹 Extends PostgreSQL’s analytical capabilities.
@p99conf @ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

Can PostgreSQL go beyond transactional workloads?

Daniel Seybold shares key insights:
🔹 DuckDB brings columnar analytics to PostgreSQL.
🔹 Citus enables distributed columnar storage.
🔹 Performance varies across analytical workloads.
@P99CONF @ScyllaDB #ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

Can smarter buffer management improve I/O efficiency?
@WitoldFormanski explores buffer rings and kernel-managed buffer selection.
🔍 Key takeaway: Buffer reuse reduces memory allocation overhead and supports concurrent I/O more efficiently.

@P99CONF @ScyllaDB #ScyllaDB #P99CONF

[Mohammed Mudasser @Md_mudasser]

🚀 What happens when networking becomes a bottleneck?

At #P99CONF 2026, @MarcinSzopa explores the asymmetric io_uring backend for Seastar.

💡 Key takeaway: Profiling CPU overhead and identifying I/O bottlenecks are essential to improving throughput.
@P99CONF @ScyllaDB #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

How can smarter I/O improve system performance?

At #P99CONF 2026, @JakubCzyszczon shares insights on dedicated networking cores and io_uring.

💡 Key takeaway: Reducing CPU contention and optimizing interrupt handling can improve I/O efficiency.

@P99CONF @ScyllaDB #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

💡💡 Key takeaways from @Daniel_Sedlak98 's #P99CONF 2026 session:

🔹 bpftrace helps uncover hidden kernel bottlenecks.
🔹 Memory-reclaim failures can severely impact TCP throughput.
🔹 Root-cause debugging can lead to Linux kernel improvements.
@ScyllaDB #P99CONF #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

🔍 When network performance suddenly collapses, the root cause may be deep inside the Linux kernel.

At #P99CONF 2026, @Daniel_Sedlak98 explores how debugging Nginx throughput revealed a TCP memory-pressure issue and led to a kernel contribution.

@ScyllaDB #P99CONF #ScyllaDB

[Mohammed Mudasser @Md_mudasser]

💡 Key takeaways from @elvisnchidera #P99CONF session:

HTTP/3 can reduce connection handshake latency and network failures, but packet loss and congestion control can impact throughput. Always validate performance with real-world A/B tests!

Learning from @ScyllaDB.

#ScyllaDB