Ask HN: When did GitHub became so slow? | Hacker News Reader

Ask HN: When did GitHub became so slow?

2 pointslinkdd7 comments

I have 1Gbps speed in download. Every single website loads instantly, I can have multiple big downloads in parallel without a single slow down.

Yet, it's been a few months, the Github website is really slow. It takes between 30s to sometimes 2-3 minutes!! (I'm not exaggerating) to load. And I'm not talking about "the webpage loads instantly then I have to wait for the actual content to show up", the webpage itself is extremely slow (by "webpage", I'm talking about that first HTTP request the browser sends to get the initial HTML/JS/etc).

Is it only me?

View on HN

I have not experienced this but 99% of my traffic to them is cloning repositories. Have you inspected your traffic in the web console of your browser? There is a network view that shows a waterfall of objects and how long it takes to get each one. I believe it also should DNS timing. It may be worth looking into that if you have time. If so I would start by clearing cache, closing the browser then opening and starting the web console in the network view so you get uncached data. Another step may be to disable all addons somewhere in your debugging as they can add red herrings.

It's always the first request in the waterfall that take the longest. All other requests are well under 100ms, except a few JS scripts that can go to 1s of loading.

I don't have any addon, I run a vanilla chrome+adblock.

Is suspected the DNS as well, but `nslookup` is instantaneous most of the time.

You mentioned adblock but do you also have antivirus? Some of them act as proxies and will look something up on the first request. Beyond that there are a million little things that can cause the first lookup to be slow. Overloaded firewall, edge router trying to purge stateful sessions, link duplex mismatches, disk cache enabled and disk being slow, MSS/MTU too high when using a VPN so the OS retrying with a smaller segment tcpdump would show this, initial window size too big microsoft will block this from people tweaking for high throughput. Try viewing the page with curl and see if that is slow. I would just be guessing at this point without knowing everything involved in the setup. If it's a network issue then Wireshark [1] may be able to highlight some issues. Also try another machine on your network and then take that machine to someone elses network like a coffee shop or bookstore that has public wifi.

[1] - https://www.wireshark.org/

Those would not explain why only Github is impacted, except maybe the antivirus one. All machines on my network are impacted as well. I'm using BitDefender as an antivirus on all of them, I might try to disable it for Github and see if it fixes the issue.

The only oddity I have noticed with Github is that if a network stack is optimized for high throughput large initial window size they seem to choke on it but you would know if you tuned for that on your router or each of your machines so it's probably not that. Have you by chance closed your browser, started tcpdump then fed the cap file into Wireshark and run the Analyze -> Expert Information tool? What OS are you using?

If Linux, as root:

    tcpdump -i any -p -NNnn -s0 -v -c10000 -w /dev/shm/p443.cap port 443

all interfaces, not promisc, disable dns, full packet size, verbos, up to 10k packets, write to a ram disk to avoid dropping, port 443 both tcp and udp just in case we try quic and time out when that fails.

I don’t have that experience. It seems to load and display content reasonably fast to me.

When it became full of AI slop and bootcamp graduates learning Python in 3 days.

Reply on news.ycombinator.com