Go programs don't just use less memory, the language makes it possible to control memory allocations.
With the JVM-based languages I'm familiar with, it is difficult to control the memory allocation characteristics of you program, and typically the further the language deviates from the JVM's model (eg, Clojure and Scala) the more garbage must be generated (usually in doing= things like runtime type reflection).
All these allocations put pressure on the garbage collector and, fortunately, the JVM provides a few sophisticated garbage collectors. But even so, I am aware of some high profile companies whose JVM-based servers spend upwards of 80% of their CPU time in the garbage collector. It's just crazy.
On the other hand, the Go language and - critically - its libraries were designed to give the programmer control over the allocation patterns of the program. This creates less work for the GC, and leads to leaner services with predictable performance characteristics.
Go has other advantages over the JVM as a platform, but this is the critical one IMO.
Go's memory model has similarities to C. For instance you can use the heap or the stack. Go figures out which one an allocation should live in by doing escape analysis. If it never leaves the function it goes on the stack. If it does it goes on the heap. GC ensures anything on the heap gets cleaned up. It's not an API.
Both the JVM and the CLR do this. This is not a Go technology.
In fact, in certain cases the JVM can detect type information and remove the allocation entirely by hoisting integer fields and the like into registers.
The programmer explicitly specifies when to use pointers. Whenever pointers are not used, the object is stored in-place (for example, on the stack, in the array or slice, or in the struct).
For instance, in Go everything is passed by value. If you want to pass a reference, you must pass a reference type (like a pointer). This means that you can pass around structs without having to put them on the heap. In JVM-land, nearly everything is passed by reference, and the data must go on the heap.
Relatedly, in Go if you allocate an array of n values of type T, the size of that array in memory is n * sizeof(T), and the items are laid out sequentially in memory. There are no other bookkeeping data structures and no indirection.
JGit struggles with not having an efficient way to represent a SHA-1.
C can just say "unsigned char[20]" and have it inline into the
container's memory allocation. A byte[20] in Java will cost an
*additional* 16 bytes of memory, and be slower to access because
the bytes themselves are in a different area of memory from the
container object. We try to work around it by converting from a
byte[20] to 5 ints, but that costs us machine instructions.
In Go, you would just write [20]byte. If you wanted a struct with a field of [20]byte you can do so, and all it costs is the 20 bytes.
The only reason you hear less "crazy" Go GC stories is because Go has never been tested as much as the JVM under different application requirements. And if high-profile companies have servers that spend 80% of the time in GC -- well, they must have really neglected their code. That's a result of poor software maintenance, and I doubt any language could help them.
> The only reason you hear less "crazy" Go GC stories is because Go has never been tested as much as the JVM under different application requirements.
It is definitely true that the JVM had been thoroughly road tested compared to Go, but your conclusion is unsound. Go was designed to specifically avoid this pitfall.
The 32-bit GC issue is an unfortunate shortcoming of the GC implementation, which will be resolved in Go 1.1 early this year. The issue was merely an artifact of dead simple conservative GC implementation.
OTOH, the memory model of the JVM is intractably linked with its design. Software that runs on the JVM must generate garbage, and so the JVM must include a highly sophisticated GC.
Remember how appallingly bad the JVM was when it was just a few years old? It didn't come close to where we are with Go today. We have only scratched the surface with GC (and other optimizations) for Go.
No... It's because Java totally sucks when objects go from Eden to Survivor and then needs to be GC'ed or from Eden to Survivor to Tenured and then needs to be GC'ed.
That's the very reason why some trading companies who still insist on using Java do use methods where as few objects as possible are ever created and went on to invent things like the LMAX disruptor "pattern" (more like an anti-pattern, where the goal is to be as close to the metal as possible, by allocating gigantic primitive arrays... Quite the contrary of your beloved GC dogma, where you think that creating objects as no impact and that slow GCs are necessarily the programmer's fault).
I realize from your various post that you did really drink the Java cool-aid and that no matter what's going to be said you'll always come touting Java as the best language out there.
I know Java well enough to criticize it.
I also have the impression that Go is used inside Google on some impressive tasks so I wouldn't be to quick to dismiss it.
I'm not dismissing Go, and I'm not praising Java, and I've even admitted on this very thread that Go is better than Java, so there's really no need in trying to convince me of anything. All I'm saying is that Go isn't good enough -- if I really want fast development and expressivity, there are languages more expressive than Go (more so than Go is to Java), and if I really need top-notch performance, then Java is OK and comes with some absolutely useful advantages that Go doesn't offer. I see excellent use cases for Python, for Ruby, for Clojure, for Java and for C/C++. I just don't see compelling use cases for Go. I'm not dismissing it. It's really nice. I just think it's underwhelming.
And, BTW, if you think Go has some magic GC secrets that will make all (or even some) of the JVM GC problems go away -- I believe you're mistaken.
I'm not sure what "impressive tasks" you mean, but I don't think that's true. I heard of people using Go, true, but the perf-heavy backends are all in C++.
With the JVM-based languages I'm familiar with, it is difficult to control the memory allocation characteristics of you program, and typically the further the language deviates from the JVM's model (eg, Clojure and Scala) the more garbage must be generated (usually in doing= things like runtime type reflection).
All these allocations put pressure on the garbage collector and, fortunately, the JVM provides a few sophisticated garbage collectors. But even so, I am aware of some high profile companies whose JVM-based servers spend upwards of 80% of their CPU time in the garbage collector. It's just crazy.
On the other hand, the Go language and - critically - its libraries were designed to give the programmer control over the allocation patterns of the program. This creates less work for the GC, and leads to leaner services with predictable performance characteristics.
Go has other advantages over the JVM as a platform, but this is the critical one IMO.