Seminarium: Systemy Rozproszone
24 października 2024 12:15, sala 4070
Kacper Sołtysiak, Jakub Panasiuk



PeRF: Preemption-enabled RDMA Framework



Remote Direct Memory Access (RDMA) enables network-intensive applications to achieve high performance by enabling zero-copy memory operations, bypassing the kernel layer. It is gaining popularity in modern computing industry, one example being adoption of RDMA by Azure for their storage access.

When it comes to enabling RDMA for cloud customers to accelerate their applications, there are problems. Modern cloud providers, to maximize their profits and resource usage, colocate virtual machines of different users (tenants) on same physical servers. This, in turn, poses challenges, such as performance isolation to prevent problem of 'noisy neighbours', whereby some customer might end up exhausting the physical resources of the underlying machine.

Because RDMA falls short in terms of performance isolation due to its inherent kernel-bypassing nature, the problem cannot be easily avoided in cloud environments.

There have been some approaches to this problem, both hardware- and software-based, each coming with non-trivial limitations either on the feasibility or performance part. Authors of the paper propose new software-based approach - PeRF - which, according to their evaluations, offers paramount performance to the previous approaches, with nearly no overhead to bare-metal RDMA.

Zapraszam,
Kacper Sołtysiak



Bibliografia:





C++ memory safety proposal



Over the past two years, the United States Government has been issuing warnings about memory-unsafe programming languages with increasing urgency. Much of the country’s critical infrastructure relies on software written in C and C++, languages which are very memory unsafe, leaving these systems more vulnerable to exploits by adversaries. The government papers are backed by industry research. Microsoft’s bug telemetry reveals that 70% of its vulnerabilities would be stopped by memory safe languages. Google’s research finds 68% of 0day exploits are related to memory corruption.

In this talk we will explore a C++ proposal that has been gathering attention, since its publication a month ago, to make C++ more memory safe. We will look at some memory leaks that could have been stopped by it and see if it can have a real possibility to be merged into C++ and if it does, will it truly solve all existing memory safety problems that the language has.

Zapraszam,
Jakub Panasiuk



Bibliografia: