Microsoft Bringing eBPF Support To Windows

[phoronix]

Phoronix: Microsoft Bringing eBPF Support To Windows

eBPF has been one of the greatest Linux kernel innovations of the past decade and now Microsoft has decided to bring this "revolutionary technology" to Windows Server and Windows 10...

https://www.phoronix.com/scan.php?page=news_item&px=Microsoft-Windows-eBPF

[oiaohm]

This could end up taking a horrible turn for the worst for Microsoft. There are a stack of patents registered by Intel and others covering eBPF in kernel mode that would require GPL license to use.

[JackLilhammers]

This could end up taking a horrible turn for the worst for Microsoft. There are a stack of patents registered by Intel and others covering eBPF in kernel mode that would require GPL license to use.

I might be wrong, but I get the impression that Microsoft has enough lawyers to make an informed decision
(and if they screw up, too bad for them...)

[uid313]

Good that Microsoft just port eBPF support directly to Windows without doing some NIH syndrome.

[Volta]

M$ trying to catch up by copying Linux technologies? Good. Ironically BPF comes from Berkley, but it's Linux not BSD having the biggest benefits from its eBPF implementation. There's still very long way for MS if they want to catch up and I bet they'll rather die in the middle of the road (maybe except for smaller servers):

JA: I found this description of pathname lookup, and it's more complex than I expected. What makes the Linux implementation so much better than what is done in other operating systems? And what do you mean by "better"?

Torvalds:
So one of the main things the VFS layer does is really handle all the locking and caching of pathname components, and handle all the serialization and the mount point traversal, and do it all with mostly lock-free algorithms (RCU), but also with some really clever lock-like things (the Linux kernel "lockref" lock is a very special "spinlock with reference count" which was literally designed for the dcache caching, and it's basically a specialized lock-aware reference count that can do lock elision for certain common situations).

End result: the low-level file systems still need to do the actual lookup for things that aren't cached, but they don't need to worry about caching and all the coherency rules and the atomicity rules that go along with pathname lookups. The VFS handles all that for them.

And it all outperforms anything any other operating system has done, while basically scaling perfectly to even machines with thousands of CPU's. And it does that even when those machines all end up touching the same directories (because things like the root directory, or a project home directory, are things that even heavily threaded applications all touch at the same time, and that don't get distributed to any kind of per-thread behavior).

So it's not just "better", it's "Better" with a capital 'B'. Nothing else out there comes even close. The Linux dcache is simply in a class all its own.

https://www.tag1consulting.com/blog/interview-linus-torvalds-linux-and-git

[Charlie68]

It would also be nice if the opposite happened every now and then, for example the possibility of using Microsoft technologies in a Linux environment, especially for what concerns us desktop users.

[M@yeulC]

It would also be nice if the opposite happened every now and then, for example the possibility of using Microsoft technologies in a Linux environment, especially for what concerns us desktop users.

Hey, it does seem to happen more often, like with exFAT recently. Microsoft could be more of a motive force, though.

[M@yeulC]

This could end up taking a horrible turn for the worst for Microsoft. There are a stack of patents registered by Intel and others covering eBPF in kernel mode that would require GPL license to use.

Isn't everyone a member of the OIN? https://openinventionnetwork.com/com...-alphabetical/

[kiffmet]

Could someone briefly explain to me what makes eBPF special, how it differs from other virtual machines (i.e .net, java) and what would be the main use case for it?

[jntesteves]

Isn't everyone a member of the OIN? https://openinventionnetwork.com/com...-alphabetical/

No. Only software companies. Hardware companies hate cooperation. They love patents and will exercise their "right" to take other companies' money at the slightest opportunity.