1 |
On Mon, Jan 18, 2010 at 11:50:55AM +0000, Stroller wrote: |
2 |
> Hi there, |
3 |
> |
4 |
> Yesterday I reseated the network cable between my server cupboard and my |
5 |
> desk, and it now lights up on the switch by my desk as gigabit. But a |
6 |
> file-transfer today is slower than I might have hoped. |
7 |
> |
8 |
> I'm not ruling out the cable, because it's pretty beat up (but the |
9 |
> switch *is* lighting up as 1000), but how do I determine, please, that |
10 |
> the Linux server at the other end is recognising the NIC and negotiating |
11 |
> as gigabit speeds? |
12 |
|
13 |
mii-tool (net-tools) or ethtool should be able to tell you that |
14 |
|
15 |
|
16 |
> The hard-drives on the server are using an older PCI SATA card, and the |
17 |
> NIC is also PCI. But I would have expected it to be a bit faster than |
18 |
> 100Mbps. |
19 |
> |
20 |
> Any estimates over what kind of speed I should be seeing for large |
21 |
> file-transfers over Samba? Wildly ball-park is fine - I wouldn't expect a |
22 |
> 10x speed increase, but maybe 2x or 3x - 4x would be great! |
23 |
|
24 |
don't know about samba, but with scp or nfs I can get about 20MByte/s which is |
25 |
the speed of my disk (and for scp almost what my cpu can manage ;) |
26 |
scp-ing /dev/zero gets me something short of 30MBye/s but that is |
27 |
because my CPU cannot manage more ;) |
28 |
|
29 |
You can see an estimate of your "raw" speed between the two machines by |
30 |
running |
31 |
nc -l -p 7777 | pv >/dev/null |
32 |
on one computer and |
33 |
pv /dev/zero | nc OTHER_COMPUTER 7777 |
34 |
on the other. I don't have a 1gbit switch here right now, so can't give |
35 |
you an estimate (with two notebooks connected directly by cable I just |
36 |
got 100MByte/s, which is near enough to the theoretical maximum ;) |
37 |
(pv is like cat, but displays a progressbar with act. speed, sys-apps/pv) |
38 |
|
39 |
you can also try netperf for more precise benchmarks |
40 |
|
41 |
> |
42 |
> I'll be testing between my Macs (both on the desktop switch, ruling out |
43 |
> both the Linux box and the suspicious cable) later today, I'd just like |
44 |
> some ideas of where I should be starting from. |
45 |
> |
46 |
> Right now I'm seeing 10 gigs of .mp4 files (1gb - 2gb per video file) |
47 |
> taking about an hour - that's about what I'd expect from old 100Mbps |
48 |
> networking, not this shiny new stuff. |
49 |
|
50 |
hmm, that seems a bit low even for 100mbit, I have usually no problem getting |
51 |
cca 10 MByte/s with 100mbit switches (without other traffic), though I |
52 |
use either nfs or scp |
53 |
the only time I remember using samba was with a winxp server, which |
54 |
didn't go above 1MB/s, but I suspect that the problem was either on the |
55 |
win side or some misunderstanding between win and linux ;) |
56 |
|
57 |
> |
58 |
> I'm not seeing any difference commenting & uncommenting "aio read size = |
59 |
> 1, aio write size = 1" (separate lines) from /etc/samba/smb.conf and |
60 |
> then running `/etc/init.d/samba reload`, but maybe I shouldn't expect |
61 |
> that to make any difference on an existing transfer. I just don't want |
62 |
> to interfere with this right now - I just want to copy as much as |
63 |
> possible on to my laptop before I go out, and I'll take a look at this |
64 |
> performance issue when I get home. |
65 |
> |
66 |
> Thanks in advance for any suggestions or pointers, |
67 |
> |
68 |
> Stroller. |
69 |
> |
70 |
|
71 |
|
72 |
|
73 |
yoyo |