performance

Customize the compilation process with Clang: Making compromises

Customize the compilation process with Clang: Making compromises

In this two-part series, we’re looking at the Clang compiler and various ways of customizing the compilation process. These articles are an expanded version of the presentation, called Merci le Compilo, which was given at CPPP in June.

In part one, we looked at specific options for customization. And, in this article, we’ll look at some examples of compromises and tradeoffs involved in different approaches.

Continue reading “Customize the compilation process with Clang: Making compromises”

Share
Customize the compilation process with Clang: Optimization options

Customize the compilation process with Clang: Optimization options

When using C++, developers generally aim to keep a high level of abstraction without sacrificing performance. That’s the famous motto “costless abstractions.” Yet the C++ language actually doesn’t give a lot of guarantees to developers in terms of performance. You can have the guarantee of copy-elision or compile-time evaluation, but key optimizations like inlining, unrolling, constant propagation or, dare I say, tail call elimination are subject to the goodwill of the standard’s best friend: the compiler.

This article focuses on the Clang compiler and the various flags it offers to customize the compilation process. I’ve tried to keep this from being a boring list, and it certainly is not an exhaustive one.

Continue reading “Customize the compilation process with Clang: Optimization options”

Share
Probing golang runtime using SystemTap

Probing golang runtime using SystemTap

I recently saw an article from Uber Engineering describing an issue they were having with an increase in latency. The Uber engineers suspected that their code was running out of stack space causing the golang runtime to issue a stack growth, which would introduce additional latency due to memory allocation and copying. The engineers ended up modifying the golang runtime with additional instrumentation to report these stack growths to confirm their suspicions. This situation is a perfect example of where SystemTap could have been used.

SystemTap is a tool that can be used to perform live analysis of a running program. It is able to interrupt normal control flow and execute code specified by a SystemTap script, which can allow users to temporarily modify a running program without having to change the source and recompile.

Continue reading “Probing golang runtime using SystemTap”

Share
How to use the Linux perf tool to count software events

How to use the Linux perf tool to count software events

The Linux perf tool was originally written to allow access to the performance monitoring hardware that counts hardware events, such as instructions executed, processor cycles, and cache misses. However, it can also be used to count software events, which can be useful in gauging how frequently some part of the system software is executed.

Recently someone at Red Hat asked whether there was a way to get a count of system calls being executed on the system. The kernel has a predefined software trace point, raw_syscalls:sys_enter, which collects that exact information; it counts each time a system call is made. To use the trace point events, the perf command needs to be run as root.

Continue reading “How to use the Linux perf tool to count software events”

Share
Speed up SystemTap scripts with statistical aggregates

Speed up SystemTap scripts with statistical aggregates

A common question that SystemTap can be used to answer involves how often particular events occur on the system. Some events, such as system calls, can happen frequently and the goal is to make the SystemTap script as efficient as possible.

Using the statistical aggregate in the place of regular integers is one way to improve the performance of SystemTap scripts. The statistical aggregates record data on a per-CPU basis to reduce the amount of coordination required between processors, allowing information to be recorded with less overhead. In this article, I’ll show an example of how to reduce overhead in SystemTap scripts.

Continue reading “Speed up SystemTap scripts with statistical aggregates”

Share
Speed up SystemTap script monitoring of system calls

Speed up SystemTap script monitoring of system calls

SystemTap has extensive libraries called tapsets that allow developers to instrument various aspects of the kernel’s operation. SystemTap allows the use of wildcards to instrument multiple locations in particular subsystems.  SystemTap has to perform a significant amount of work to create instrumentation for each of the places being probed.  This overhead is particularly apparent when using the wildcards for the system call tapset that contains hundreds of entries (syscall.* and syscall.*.return). For some subsets of data collection, replacing the wildcard-matched syscall probes in SystemTap scripts with the kernel.trace("sys_enter")  and the kernel.trace("sys_exit") probe will produce smaller instrumentation modules that compile and start up more quickly. In this article, I’ll show a few examples of how this works.

Continue reading “Speed up SystemTap script monitoring of system calls”

Share
How data layout affects memory performance

How data layout affects memory performance

The mental model most people have of how computer memory (aka Random Access Memory or RAM) operates is inaccurate. The assumption that any access to any byte in memory has the same low cost does not hold on modern processors. In this article, I’ll explain what developers need to know about modern memory and how data layout can affect performance.

Current memory is starting to look more like an extremely fast block storage device. Rather than reading or writing individual bytes, the processor is reading or writing groups of bytes that fill a cache line (commonly 32 to 128 bytes in size). An access to memory requires well over a hundred clock cycles, two orders of magnitude slower than executing an instruction on the processor. Thus, programmers might reconsider the data structures used in their program if they are interested in obtaining better performance.

Continue reading “How data layout affects memory performance”

Share
Register Transfer Language for CRuby

Register Transfer Language for CRuby

For the last two years, I have been trying to improve CRuby performance. I have been working simultaneously on two major fronts: introducing register transfer language (RTL) for the CRuby virtual machine (VM) and just-in-time (JIT) compilation. For background on the goal of having Ruby 3 be 3 times faster than version 2 (3X3), see my previous article, “Towards the Ruby 3×3 Performance Goal“.

The JIT project (MJIT) is advancing successfully. The JIT approach and engine I proposed and implemented has been adopted by the CRuby community. Takashi Kokubun hardened the code and adapted it to the current CRuby stack machine and recently MJIT became an experimental feature of the CRuby 2.6 release.

Introducing a Register Transfer Language (RTL) to the CRuby VM turned out to be an even harder task than introducing the initial JIT compiler. The required changes to the VM are far more invasive than the ones needed for the JIT compiler.

This article describes the advantages and disadvantages of RTL for CRuby.

Continue reading “Register Transfer Language for CRuby”

Share
Algorithms != Programs and Programs are not “One size fits all”

Algorithms != Programs and Programs are not “One size fits all”

You’ve probably been taught that picking an algorithm that has the best Big-O asymptotic cost will yield the best performance. You might be surprised to find that on current hardware, this isn’t always the case. Much of algorithmic analysis assumes very simple costs where the order of operations doesn’t matter. Memory access times are assumed to be the same. However, the difference between a cache hit (a few processor clock cycles) and a cache miss that requires access to main memory (a couple hundred cycles) is immense.

This article series is the result of the authors (William Cohen and Ben Woodard) discussion that there is a disconnect on the typical ideas of algorithm efficiency taught in computer science and computer engineering versus what is currently encountered in actual computer systems.

Continue reading “Algorithms != Programs and Programs are not “One size fits all””

Share