Extremely slow performance on Fedora 10

Asked by Cam

I am very new to Fedora 10 but have been tinkering. I have a setup now that has all the updates and my graphics card drivers. After hashing LinuxDC++ and connecting to a Gigabit server at a LAN Party I was getting extremely slow download speeds compared to everyone else. Around 8MB/s usually. Any suggestions?

Question information

English Edit question
LinuxDC++ Edit question
No assignee Edit question
Last query:
Last reply:
Revision history for this message
steakunderscore (steakunderscore) said :

To be honest I'm not convinced this is a problem with linuxDC++. You might be aware of this, hence you pointing out you were connected by gigabit, but 8MB/s is quite quick.
Take a 100mb/s (Note the lower case 'b') divide that by 8 to get MB/s. You get 12.5MB/s theoretical maximum speed. After you take into account things like packet headers you are down to about 10MB/s.

I hope I haven't patronised you, just want to make sure you know some of the basics.

Revision history for this message
Cam (ckeech) said :

Thanks steakunderscore. You haven't patronised. I didn't elaborate enough. I am also not at all convinced its a problem with LinuxDC++ but I figured it was possibly an incorrect setting. If I have a Gigabit connection then I have a theoretical maximum of 125MB/s connection speed. As you said it would almost never reach that, but for arguments sake I would assume that, say 50MB/s or at least 25MB/s would be perfectly reasonable considering that there are multiple connections as well, as in downloading different things from about 15 people and in some cases there were multiple people with the same file. My friend next to me was running DC++ on Windows XP and was getting considerably higher speeds. Around the 50MB/s mark.

I am also aware that there are many variables that affect this speed. I found at that the hard drive I was downloading too was connected to a cheap and nasty PCI card that was really slow. After changing to a SATAII connected drive it increased the speeds dramatically (from KB/s to MB/s). I guess I'm expecting to reach similar speeds to my friends' Windows computers, Well I was assuming that Linux would be more resource efficient than Windows so I was expecting to exceed my friends computers.

Revision history for this message
Pseudonym (404emailnotfound) said :

You might get somewhat better performance from disabling safe and compressed transfers. Also, netcat and pv could be useful in measuring raw throughput between two machines (obviously requires POSIX or Cygwin). Something like the following:
comp1$ nc -l -p 6000 < /dev/zero
comp2$ nc $comp1_ip_address 6000 | pv > /dev/null

Revision history for this message
box2 (linuxdcpp) said :

I've found that invariably transfers between Windows and Linux run on the less than efficient side using most even well established protocols. For example, running a Samba server, my Windows computer could only download files across my LAN at 1.5MB/s, using FTP sometimes it would spike to 3MB/s then back to something lower. When using an all Linux network (as I do now, sharing with NFS) I max out my 100mbit LAN regularly.

On a side note, also take into consideration packet collisions on the network. If 10 people are using the same gigabit switch/router all grabbing from a single share, not only will none of them be able to get maximum network speed, but severe degradation will occur on account of packet handling by said switch/router. If it was a hub (without switch), then it doesn't matter if it was only a single person serving, as all communication will travel the same backbone through the hub and all communication will slam together and the endless amount of packet re-requests will look something akin to hard drive swap hell (when you run out of system RAM).

On another aside, you are right in that IDE drives are damn slow, and in general much slower than LAN. SATA II however, as you mention, is finally up to the task of maxing out a single way transfer across gigabit LAN. For reference:


Yet keep in mind drive speeds vary greatly depending on the type of reads/writes and where on the platter they are reading/writing too (the inner most rings on the drive will by transfer far slower than the raw bus speed of the controller for the device). Not to mention brands of drives handle things very differently, and some will be better than others in certain types of transfer from certain sectors and blah blah.

Can you help with this problem?

Provide an answer of your own, or ask Cam for more information if necessary.

To post a message you must log in.