majke

12 ms·

121.
by majke 3y ago
However this is not the case for ipv6. Technically you can use only ::1, unless you do Ipv6 FREEBIND
122.
by majke 3y ago
Always use Unix Domain sockets if you can. There are at least three concerns with TCP. First, local port numbers are a limited resource. https://blog.cloudflare.com/how-to-stop-running-out-of-ephem... https://blo
123.
by majke 3y ago
oh, oh. I know what. I don't know much about the context, but I remember that datetime parsing code in Firefox was one of the messiest parsers around: https://searchfox.org/mozilla-central/source/nsprpub/
124.
by majke 3y ago
Yup! The details of Linux networking API are often surprising. I think if it's possible to change the ping algo to reduce the potential for wall clock change affecting the measurement. For example, instead of using SO_TIMESTAMP to meas
125.
by majke 3y ago
more like "inflight"
126.
by majke 3y ago
Thanks Dan for reposting :) If you want more of this stuff, follow my twitter: https://twitter.com/majek04/status/1687023191506530304 Lately it seems I'm spending way too much time on ping. This is from last
127.
by majke 3y ago
What a blast from the past. I remember using pegasus back in the days. I loved it, not sure why I stopped using it. I guess gmail search function?
128.
by majke 3y ago
I used vx32 as a technical demo before workers came to be. It was super useful. However, vx32 itself is very much not production ready. There are issues with instruction decoding, floating point, and interrupt handling. (it uses itimer as a
129.
The day my ICMP ping took countermeasures(blog.cloudflare.com)
3 points by majke 3y ago|0 comments
130.
by majke 3y ago
Hey, yeah, they are using CLOCK_MONOTONIC for the event loop exactly for that. I think it might be possible to fix the SO_TIMESTAMP wall-clock problem by using it not to measure duration between ping send and ping receive, but to measure ke
131.
by majke 3y ago
Cool. Great, well written and well researched. I like the form - prose mixed with source links. Reads like a story. When I researched similar subject in past, though, I focused on the user-exposed toggles/counters [0]. It's very h
132.
City of love and war: How a front-line city became Ukraine’s romantic capital(economist.com)
2 points by majke 3y ago|1 comments
133.
by majke 3y ago
Hermaphrodite Connector????
134.
by majke 4y ago
There are probably a couple of independent things: (1) HW GRO, most often LRO (TCP) (2) SW GRO aka GRO Hardware GRO was typically called LRO for TCP and is often disabled in practice. One reason is XDP, with XDP some offloads need to stay o
135.
by majke 4y ago
Okay, as far as I understand this writeup. There are two sides, userspace UDP socket to receive wg packets on. Then the tap file descriptor to receive unencrytped packets from the host OS. To speed up the userspace UDP socket it's desi
136.
Tech-noir: technology as a destructive and dystopian force(en.wikipedia.org)
2 points by majke 4y ago|2 comments
137.
by majke 4y ago
Thanks for reminder. This was a blog post after we hit the connectx() troubles :) Tubular/sk_lookup is about ingress. This blog post is about connected sockets on “egress”. Actually, bpf sk lookup could be used on egress, but its not q
138.
by majke 4y ago
Let me try. You could expect that on linux host you can have 64K concurrent connections to single target ip. If you had two target ips you could expect 128k concurrent conns. This is not how it works. Especially when doing bind-before-conne
139.
by majke 4y ago
Vincent Bernat's writings are the best. On my side, our experience with flowspec was not perfect, although arguably it was 10 years ago: https://blog.cloudflare.com/todays-outage-post-mortem-82515/ https:/&#
140.
Ralf Brown's Interrupt List(en.wikipedia.org)
136 points by majke 4y ago|65 comments
141.
by majke 4y ago
Back in 2019 I wrote about "splicing" in the context of networking: https://blog.cloudflare.com/sockmap-tcp-splicing-of-the-futu... Many of the mentioned bugs are fixed by now, though arguably I'm focusing mo
142.
by majke 4y ago
I'm a pragmatist and there are couple of points in this article that give me allergic reaction. > You want individual messages instead of a byte-stream? TCP has an option for that. That links to. https://book.systemsappro
143.
Features and flaws of speedtest.net, ookla and Cloudflare(blog.cerowrt.org)
1 points by majke 4y ago|0 comments
144.
by majke 4y ago
https://blog.cloudflare.com/accelerating-udp-packet-transmis...
145.
by majke 4y ago
ChatGPT definitely improved over last two weeks. I'd say it has just crossed "The enterprise central computer" intelligence threshold. It's able to express facts its aware of in a concise and reasonable way. You can hold
146.
by majke 4y ago
I can chime in with some optimizations (linux). For normal UDP sockets UDP_GRO and UDP_SEGMENT can be even faster than sendmmsg/recvmmsg. In Gvisor they decided that read/write from tun is slow so they did PACKET_MMAP on raw socke
147.
by majke 4y ago
There is more: * pcp https://datatracker.ietf.org/doc/html/rfc6887 * ds-lite port allocation mode https://support.huawei.com/enterprise/en/doc/EDOC1100125799/...
148.
by majke 4y ago
Hi there! In the article I only briefly mentioned IPv6, it was long enough. Indeed, we do happy-eyeballs where we can and _strongly_ prefer IPv6 when origin host resolves AAAA. However, we still need IPv4, since there is still a big chunk o
149.
by majke 4y ago
"Overall, it seems like they are treating ip addresses as data essentially" Spot on! In past: * /24 per datacenter (BGP), /32 per server (local network) (all 64K ports) New: * /24 per continent (group of colos), &#x
150.
by majke 4y ago
Correct. From a single place, you're likely to BGP-reach one Cloudflare location, and it doesn't change often.

More ›