#1844 new defect

slow download speeds

Reported by: l3iggs Owned by: davidsarah
Priority: normal Milestone: undecided
Component: code Version: 1.9.2
Keywords: performance download Cc:
Launchpad Bug:

Description

I have 2 storage servers and one client machine.
storage-serverA = 4x1.6 GHz, 3GB of ram, on home lan
storage-serverB = 2x2.3 Ghz, 31GB of ram, remote locaction
client = 4x2.4 GHz, 4 GB of ram, on home lan
I have
k=1
N=2
h=2
When I download a 104MB file from my cluster using
wget http://localhost:$PORT/uri/$CAP
wget reports that file transfer takes place at ~1.5M/s and completes in 67 seconds
In this example, storage-serverA (on my lan) transfers the entire file (see attached file dlStatus.html)
During the transfer tahoe consumes 76% of one core of the CPU on storage-serverA. And 16% of one core of the CPU on client.

The same file transferred using
scp user@storage-serverA:/file[[BR]] takes place at ~12.3 MB/s and completes in 8 seconds.
During this transfer, sshd consumes 103% of one core of the CPU on storage-serverA. And ssh consumes 22% of one core of the CPU on client.

tahoe-lafs seems to be ~8x slower on a point to point transfer over my lan than ssh is.

Attachments (1)

dlStatus.html (509.1 KB) - added by l3iggs at 2012-11-06T19:16:14Z.
tahoe download status

Download all attachments as: .zip

Change History (2)

Changed at 2012-11-06T19:16:14Z by l3iggs

tahoe download status

comment:1 Changed at 2012-12-11T22:28:45Z by davidsarah

  • Component changed from unknown to code
  • Keywords download added
Note: See TracTickets for help on using tickets.