Major Network Performance Regressions In Linux
Affecting the latest Linux kernel release, Linux 3.7, are "multiple apparently unrelated network performance issues." The major network performance problems were reported by a well-known Linux kernel developer.
Willy Tarreau, a Linux kernel developer and the one that was the maintainer of the Linux 2.4 kernel series, wrote a new mailing list thread to kernel developers on Saturday that was entitled "Major network performance regression in 3.7." The problems also seem unresolved by the current Linux 3.8 kernel.
Tarreau basically is seeing multiple network performance issues in the stable 3.7 kernel. The problems are so bad he's doubting they even come from the Linux network stack. For anyone into Linux networking that want to know all of the technical details concerning these regressions, see the mailing list thread.
Among the possibilities talked about in response to these performance regressions within the Linux kernel is that interrupts are being spread on multiple CPUs and that is resulting in out-of-order problems with the Linux TCP stack and the Myri network card / driver not liking high order pages, but Willy Tarreau then said he thinks this might actually be a problem with the Linux memory allocation.
At the time of publishing the post, these Linux network performance regressions are still being investigated. Update: Since the original posting of this article, a few updates on the network problems have been posted to the kernel thread.
Willy Tarreau, a Linux kernel developer and the one that was the maintainer of the Linux 2.4 kernel series, wrote a new mailing list thread to kernel developers on Saturday that was entitled "Major network performance regression in 3.7." The problems also seem unresolved by the current Linux 3.8 kernel.
Tarreau basically is seeing multiple network performance issues in the stable 3.7 kernel. The problems are so bad he's doubting they even come from the Linux network stack. For anyone into Linux networking that want to know all of the technical details concerning these regressions, see the mailing list thread.
Among the possibilities talked about in response to these performance regressions within the Linux kernel is that interrupts are being spread on multiple CPUs and that is resulting in out-of-order problems with the Linux TCP stack and the Myri network card / driver not liking high order pages, but Willy Tarreau then said he thinks this might actually be a problem with the Linux memory allocation.
At the time of publishing the post, these Linux network performance regressions are still being investigated. Update: Since the original posting of this article, a few updates on the network problems have been posted to the kernel thread.
7 Comments