network transfer between two nodes very slow

[satra]

Version information:

server:
go-ipfs version: 0.4.6-
Repo version: 5
System version: amd64/linux
Golang version: go1.8

client:
go-ipfs version: 0.4.6-
Repo version: 5
System version: amd64/darwin
Golang version: go1.8

Type: bug

Priority: P3

Description:

on the client:

ipfs bootstrap rm --all
ipfs bootstrap add /ip4/18.13.53.53/tcp/4001/ipfs/QmSgC4gsjk3hd49Yosov2CadExkRts9dAdoEEyG2TN3jHu
ipfs ping QmSgC4gsjk3hd49Yosov2CadExkRts9dAdoEEyG2TN3jHu
ipfs get -o ants /ipfs/QmNng3MdmWZCxkBbwrxbDEoXXQ5gqvcCSuPisdAxvcRCh2

and this transfer is very slow.

the server sits on a large bandwidth connection and if i simply use rsync to download the ipfs directory it's much much faster.

is there a way to determine what the client and the server are spending their time on?

@whyrusleeping - it almost seems like the ls issue we were seeing. that it won't start retrieving anything till it does an ls and then takes forever to download the 1.5G of data.

[whyrusleeping]

@satra Hrm... This is a bit of a known problem. The dht overhead here is pretty annoying (something i'm planning on improving with: #3786).

What is the normal transfer speed between the two machines in question?

What is the latency between the two machines?

What is the actual bandwidth utilization of each machine vs how fast the transfer appears to be going?

[satra]

@whyrusleeping - sorry about the delay. had a chance to upgrade to 0.4.9 and test out latencies and bandwidth.

ping latency:

64 bytes from 18.13.53.53: icmp_seq=0 ttl=55 time=6.908 ms
64 bytes from 18.13.53.53: icmp_seq=1 ttl=55 time=7.288 ms
64 bytes from 18.13.53.53: icmp_seq=2 ttl=55 time=7.355 ms
64 bytes from 18.13.53.53: icmp_seq=3 ttl=55 time=7.412 ms

an scp command:

$ scp satra@oms.mit.edu:/path/to/file .
sub-BANDA001_task-rest_run-01_bold.nii.gz                 12%   60MB   3.7MB/s   01:57 ETA

ipfs command translates to about 20MB/minute

ipfs get -o testfs /ipfs/QmWvQ8fz9XMJoxQHV5BHKmVqYoVNWkckUxQ8LNSgEkdBgD/bin
Saving file(s) to testfs
 41.97 MB / 2.47 GB [>----------------------------------------------------]   1.66% 1h57m3s

[satra]

@whyrusleeping and @Kubuxu - any ideas on how to improve throughput?

[magik6k]

#3867 was merged so it should be better now. You can build go-ipfs from current master and see if it improves things for you.

[satra]

@magik6k - will check it out and report back the numbers.

[Kubuxu]

You may also want to start the daemon with ipfs daemon --routing=none for receiving the files. We are thinking of better providing strategies so they don't reduce speed of the download but this might be workaround for now.

[satra]

@Kubuxu - ok this is much faster. i'll keep playing with and see if things are better.

[Kamelia2000]

Hello Do you think IPIPFS is good to make a chat soluton like IM messenger like Telegram and very much fast ?

[Stebalien]

@Kamelia2000 please ask support questions on https://discuss.ipfs.io

(also, in general, please keep issue comments on-topic)