Making Unsafe safer

Overview

If you use Unsafe directly, you risk crashing the JVM.  This happens when you access a page of memory which hasn't been mapped and the result on Unix is a SIGSEG (if you access page 0) or SIGBUS (if you access another page which is not mapped)

Using MethodHandles

Wrapping Unsafe method with a MethodHandle is a possible solution.  You can add code to Method Handles to check for a zero page access. e.g. unsigned_ptr < 4096.  The reason you would add this to MethodHandle is it makes it easier to optimise away this check.

Downside of this is that

  • You have to use MethodHandles which complicates the syntax and obscures what you are really doing.
  • It doesn't work if you don't
  • It doesn't cover bus errors, nor could it as the mapping for the whole application is complex, and can change in any thread at any time.
  • Optimising away the bounds check requires some work to the optimiser which is yet to be proven.

Using Signals.

If only there was some way to do this in the hardware already, and there is.  The CPU already checks if a page you attempt to access is valid and it throws an interrupt if the page is not in cache. This interrupt is turned into a signal if the OS cannot find/create a mapping for the this cache miss.

If only there was a signal handler in the JVM already, and there is, this is what produces the crash report.

If only there was some way for an interrupt handler to trigger an error or exception back to the code which triggered it.  Like Thread.currentThread().stop(e);   (You get the idea)

Advantages
  • No additional work is required to perform the checking as it is already done by the CPU.
  • Minimal changes to the optimiser (if any)
  • Potentially work for signals produced from a range of source.
  • Using signals are a mature/old tech way of trapping runtime errors which pre-date Java.
Disadvantages
  • Single processing is likely to be a stop-the-world operation (no way to benchmark this in Java currently)
  • Even if it is not, it is likely to be much more expensive when an error is triggered.
  • You would have to change the signal handler which traditionally hasn't been changed. i.e. there is much more experience of changing the optimiser.

Possible exceptions thrown.

New exceptions could be thrown, however I suggest re-using existing exceptions.

Access to page 0 - NullPointerException

Accesses to page 0 (not just access of a NULL pointer) trigger a SIGSEG.  NPE is named after the access of a NULL pointer from C and it is perhaps more obvious to have a NPE for an access to a NULL pointer than a reference. i.e. it could have been called NullReferenceException since Java doesn't have pointers.

Invalid Access - IndexOutOfBoundsException

Other candidates include BufferUnderflowException (if you a a page short of a mapped region), BufferOverflowException (if you are a page long of a mapped region)

Something these all have in common is they are RuntimeException(s).  If a custom, more descriptive exception is raised, a RuntimeException might be consistent with existing throwables thrown.

Conclusion

A common trick to maximise performance is; don't write in Java something your system is doing for you already.  In Chronicle we use the OS to do the asynchronous persistence to disk and it is more efficient and reliable than writing the same again in Java.  Similarly, trapping and handling invalid memory access would be more efficient and more robust if the facilities provided by the CPU and OS were re-used.

Generally speaking you would re-write OS features when each OS does things differently to support cross platform compatibility, but only a minimum required to do this.  This is why Java doesn't have a thread scheduler and in relativity has little control over how threads are run.  

Virtual memory handling is so old and standard, that major platforms all work basically the same way.

Comments

Post a Comment

Popular posts from this blog

Java is Very Fast, If You Don’t Create Many Objects

Image
  You still have to watch how many objects you create. This article looks at a benchmark passing events over TCP/IP at 4 billion events per minute using the net.openhft.chronicle.wire.channel package in Chronicle Wire and why we still avoid object allocations..  One of the key optimisations is creating almost no garbage. Allocation is a very cheap operation and collection of very short-lived objects is also very cheap. Does this really make a difference? What difference does one small object per event (44 bytes) make to the performance in a throughput test where GC pauses are amortised? While allocation is as efficient as possible, it doesn’t avoid the memory pressure on the L1/L2 caches of your CPUs and when many cores are busy, they are contending for memory in the shared L3 cache.  Results Benchmark on a Ryzen 5950X with Ubuntu 22.10. JVM Vendor, Version No objects Throughput, Average Latency* One object per event Throughput, Average Latency* Azul Zulu 1.8.0_322 60.6 ...
Read more

System wide unique nanosecond timestamps

A Unique Identifier can be very useful for tracing. Those ids are even more useful when they contain a high-resolution timestamp.  Not only do they record the time of an event, but if unique can help trace events as they pass through the system. Such unique timestamps however can be expensive depending on how they are implemented.   This post explores a lightweight means of producing a unique, monotonically increasing system-wide nano-second resolution timestamp available in our open-source library. Uses for Unique Identifiers Unique identifiers can be useful to associate with a piece of information so that information can be referred to later unambiguously. This could be an event, a request, an order id, or a customer id. They can naturally be used as a primary key in a database or key/value store to retrieve that information later. One of the challenges of generating these identifiers is avoiding creating duplicates while not having an increasing cost.  You could ...
Read more

What does Chronicle Software do?

Image
Overview Chronicle Software is about simplifying fast data.  It is a suite of libraries to make it easier to write, monitor and tune data processing systems where performance and scalability are concerned. But its free, how do you make money, through support? We offer premium support . However, we often hear from users for the first time about a year after they have it in production.  Most users find they can support the software themselves.  It is only after a year or so, they have questions or concerns about where to take the product next.  In many cases, it is to a problem we have already solved and they just need to upgrade their software or use a new module we have added. As we don't know who is using most of our software, we can't advise them on how best to use our software and what updates or enhancements they would benefit from. We need users to contact us and ask questions.  We have a free forum , and we respond to 50% of questions in 2 hours...
Read more