Page MenuHomePhabricator

Access to Wikimedia sites very slow from India
Closed, DeclinedPublic

Description

Since yesterday (Feb 15), acces and edits are Wikimedia sites are very slow from India. Tested on Wikisource, Wikipedia, and Commons from latest Chrome and Firefox on Windows 7. I changed hosts to use proxies in Netherlands, and it is now much faster, so it does not seem to be an issue on my side.

Acces and edits are much faster with these settings in hosts:
91.198.174.192 fr.wikisource.org
91.198.174.202 bits.wikimedia.org
91.198.174.192 commons.wikimedia.org

Results from http://winmtr.net/download-winmtr/ :

WinMTR statistics
Host - %SentRecvBestAvrgWrstLast
192.168.0.1 - 025251121
113.20.16.1 - 025251252
113.20.19.33 - 025251241
113.20.19.21 - 025252252
182.19.18.126 - 5212012132112
No response from host - 100400000
xe-7-0-1-xcr2.lsw.cw.net - 80510150150150
No response from host - 100400000
No response from host - 100400000
No response from host - 100400000
No response from host - 100400000
No response from host - 100400000
bits-lb.eqiad.wikimedia.org - 02525206208212211
____________________________________________________________________________________

WinMTR v0.92 GPL V2 by Appnor MSP - Fully Managed Hosting & Cloud Provider

Results from pathping bits-lb.eqiad.wikimedia.org -q 10 :
C:\Users\USER>pathping bits-lb.eqiad.wikimedia.org -q 10
Tracing route to bits-lb.eqiad.wikimedia.org [208.80.154.234]
over a maximum of 30 hops:

0  USER-PC [192.168.0.100]
1  192.168.0.1
2  113.20.16.1
3  113.20.19.33
4  113.20.19.21
5  182.19.18.126
6     *        *        *

Computing statistics for 12 seconds...

Source to Here   This Node/Link

Hop RTT Lost/Sent = Pct Lost/Sent = Pct Address

0                                           USER-PC [192.168.0.100]
                              0/  10 =  0%   |
1    1ms     0/  10 =  0%     0/  10 =  0%  192.168.0.1
                              0/  10 =  0%   |
2    1ms     0/  10 =  0%     0/  10 =  0%  113.20.16.1
                              0/  10 =  0%   |
3    3ms     0/  10 =  0%     0/  10 =  0%  113.20.19.33
                              0/  10 =  0%   |
4    2ms     0/  10 =  0%     0/  10 =  0%  113.20.19.21
                              0/  10 =  0%   |
5   13ms     0/  10 =  0%     0/  10 =  0%  182.19.18.126

Trace complete.

According to my Resource Monitor, I have a bandwidth of 1 K to bits-lb.eqiad.wikimedia.org
while http://www.speakeasy.net/speedtest/ gives me:
Download Speed: 1984 kbps (248 KB/sec transfer rate) to Washington D.C.


Version: wmf-deployment
Severity: minor

Details

Reference
bz61437

Event Timeline

bzimport raised the priority of this task from to Low.Nov 22 2014, 2:58 AM
bzimport set Reference to bz61437.
bzimport added a subscriber: Unknown Object (MLST).

What are the first three octets of your IP? Are there any other users experiencing similar delays?

My IP is 103.251.218.x. I changed the importance to major. This is really a problem, and hacking /etc/hosts is not a viable solution. Thanks, Yann

My ISP is Gujarat Telelik Pvt. if it matters.

I don't think it's "major" or even "normal" until we have reports from multiple users. Have you tried contacting your ISP to report your issues?

According to http://newsroom.cisco.com/dlls/1911/prod_021511.html , this ISP dominates market in a region with hundreds millions inhabitants, let's hope they don't all go through that culprit...

(In reply to Yann Forget from comment #0)

182.19.18.126 - 5212012

13 | 21 | 12 |

No response from host - 100400

0 | 0 | 0 |

xe-7-0-1-xcr2.lsw.cw.net - 80510

150 | 150 | 150 |

Havoc begins here with Vodafone, right? As Faidon suggests, contacting the ISP may solve the issue easily if there is a bottleneck somewhere and they can bypass it; with sufficiently precise reports, I often had success simply by writing my ISP's abuse@ address (even though they fixed stuff silently), whatever circumvents the usually unhelpful customer services. You may even try the support address listed at http://www.cw.net/, can't harm. ;)

CW = Cable&Wireless, acquired by Vodafone 2 years ago and serves as their global carrier network, interconnecting all of the local Vodafones. They're exclusively a carrier, carrying Vodafone India's traffic, which in turn carries Gujarat Telelik Pvt. It doesn't make any sense to contact them as an end-user, both because they're a large-scale carrier and because you're not a customer of theirs.

I can and will contact them (as we peer with them) but not without having data that suggest a general rather than an individual problem from just one user. There's ISPs' support desks for reporting individual problems. Either us or the ISP can spot a general pattern and raise it with our respective upstreams until we reach the problematic network.

Let's leave this ticket open in case more reports come in. If not, I'd suggest closing it in e.g. 2 weeks time.

(In reply to Faidon Liambotis from comment #4)

I don't think it's "major" or even "normal" until we have reports from
multiple users. Have you tried contacting your ISP to report your issues?

As I said above. evereything is working fine *except* Wikimedia sites. And it seems to be only when accessing the eqiad servers. When changing /etc/hosts to use cache in Netherlands, it works much better (although Javascript loading is still very slow). So it doesn't make sense to contact my ISP, who will rightly send me off.

Yann: Any chance to reach out to other users in India and if anybody else sees this problem?

Yann: Any chance to reach out to other users in India and if anybody else sees this problem?

Hi, It seems that issue disappeared as it came. I may reopen this if it comes back. Thanks, Yann

Thanks. No fix involved, hence resetting to WORKSFORME.