In the past? It was a bit of a mixed bag. TCP performance and network latency in general was always worse on Windows than Linux, largely for silly reasons. Similarly, the 32-bit versions had a lot of fixed sized memory regions for things like file caching that were set in stone in the NT4 era and not changed until Windows Vista. Towards the end when all 32-bit machines had 4GB of memory, this made disk access unnecessarily slow.
There are also design space differences between Linux and Windows. On Linux a process is very lightweight, basically a fat function call. On Windows the equivalent need is typically met by RPC calls to an already running process. Threads are cheap on Windows and threading was very well supported since the 1990s. On Linux, not as well, even now. For example, even NT4 and async scatter/gather IO that SQL Server used back in v7. I believe that Linux is still struggling with this.
Conversely, Windows beats the pants off Linux in some areas, even now. Try using a Linux desktop with a flaky DNS server. Hooo-boy. Timeout city! On Windows this is almost unnoticeable, because it fails over to the secondary DNS server in milliseconds and caches all results as expected. Linux failover and caching seems to require a significant effort to set up by the users, for some mysterious reason.
Similarly, Windows SMB v3 over RDMA is crazy fast. If you have the right NIC, you can flatline a 40 Gbps pipe with a single file copy. In the Windows Explorer GUI! No special tools needed. Copy, paste, and next thing you know you're doing 5 gigabytes per second. That takes real effort to match in Linux.
Etc...
Basically, everyone is used to the warts on their preferred system, and has learned to work around them. But when exposed to something unfamiliar, all the little issues seem to be always in the way.
I would disagree. Unless you're trying to write "cross-platform" network code using select and other POSIX-style interfaces, in my experience Windows with IOCP or RIO will beat any other user mode application TCP stack for I/O rates and latency.
Vista's stack had latency issues in the beginning but Windows 8 with RIO was released a long time ago.
Note I said "user mode application". Linux has easier kernel bypass and well-supported DPDK for applications that really need it. You obviously cannot beat that with user-mode code
Where people get into trouble with it is when they try to code against it like it's another Unix variant.
https://randomascii.wordpress.com/2019/09/08/taskbar-latency...
I thought this was a particularly bizarre take, given that in the very same post the author has lamented not having access to the source code of the program he is trying to debug.
You can keep your symbol servers, I'll take the source code please.
On Linux, you might run into the problem of a distro shipping stripped binaries, which means having the source code doesn't help much in parsing a stack or profiler trace. Instead, you need to track down the debug info for the binaries as compiled and shipped by that particular distro. Even in the best case, that will be more work than Microsoft's single-source symbol server.
Stuff like "npm install", large git operations are many times slower than doing the same operation in a linux virtual machine running on that same Windows. Hard to tell if it's Windows itself, or maybe npm/git not using the "proper ways"
Other than that, Windows performance it just fine, and arguably smoother than a Linux desktop (no mouse freezing for example).
The Stockholm syndrome is a very real thing.
The "I'm very sorry" comment was meant to be strictly in relation to symbol servers where Windows is clearly ahead (although Linux may be making progress in this area).
Secondly, I think it's a very good point you raised. Sad to see the situation hasn't progressed much in the linux world, as reading about Microsoft's symbol servers, they seem incredibly useful!
I keep reading these articles about terrible quadratic complexity for things on windows, it gives me the impression that performance on windows can be pretty bad, and frankly, I don't have the best opinion of windows.
Is performance worse in general on windows? I haven't used it for a fair few years. Is it just generally a worse OS? I used to somewhat resent having to use windows when I needed to develop on it (developing for games consoles) but there were some nice parts (I have to say, visual studio is really nice as an integrated debugger, even if I prefer other editors).
I'm trying to check my biases. Is windows actually a fairly poor OS nowadays?
> If you are not on Windows – I’m very sorry.
The writer on the article definitely seems to think that it's a nice one, despite the constant issues they run into. Is this just because they've used the system so long, and have developed such expertise with it, or there are things to love that I'm missing about it?