I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with 2.4.x to copy a 36MB file - same HW) Does anybody have some hints, other than upgrade to newer kernel, since not always possible? Thanks
Marco De Maddalena wrote:
I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with 2.4.x to copy a 36MB file - same HW) Does anybody have some hints, other than upgrade to newer kernel, since not always possible?
Are you sure it is network performance, how about local disk I/O. If you're using IDE drives could be hdparm settings that are to blame. -- Simon Oliver
Yes it is network performance: I took exactly the same PC, and I did exactly the same operation. It has been verified in several circustances, with several PCs. -----Messaggio originale----- Da: Simon Oliver [mailto:simon.oliver@umist.ac.uk] Inviato: lunedi 17 marzo 2003 10.53 A: Marco De Maddalena Cc: Suse-Linux-E Oggetto: Re: [SLE] Network performances Marco De Maddalena wrote:
I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with 2.4.x to copy a 36MB file - same HW) Does anybody have some hints, other than upgrade to newer kernel, since not always possible?
Are you sure it is network performance, how about local disk I/O. If you're using IDE drives could be hdparm settings that are to blame. -- Simon Oliver
On Monday 17 March 2003 11:30, Marco De Maddalena wrote:
Yes it is network performance: I took exactly the same PC, and I did exactly the same operation. It has been verified in several circustances, with several PCs.
-----Messaggio originale----- Da: Simon Oliver [mailto:simon.oliver@umist.ac.uk] Inviato: lunedi 17 marzo 2003 10.53 A: Marco De Maddalena Cc: Suse-Linux-E Oggetto: Re: [SLE] Network performances
Marco De Maddalena wrote:
I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with
2.4.x
to copy a 36MB file - same HW) Does anybody have some hints, other than upgrade to newer kernel, since
not
always possible?
Are you sure it is network performance, how about local disk I/O. If you're using IDE drives could be hdparm settings that are to blame.
Could it be that the NICs are set to half-duplex in an otherwise full-duplex ethernet network? -- Joris Verbogt Cirrus International BV T: +31 15 2517575 F: +31 15 2517579 E: jverbogt@cirrus.nl W: http://www.cirrus.nl/ -- use Perl; #powerful programmable prestidigitation
On Monday 17 March 2003 04:39 am, Joris Verbogt wrote: <snip>
Are you sure it is network performance, how about local disk I/O. If you're using IDE drives could be hdparm settings that are to blame.
Could it be that the NICs are set to half-duplex in an otherwise full-duplex ethernet network?
-- Joris Verbogt Cirrus International BV T: +31 15 2517575 F: +31 15 2517579 E: jverbogt@cirrus.nl W: http://www.cirrus.nl/
-- use Perl; #powerful programmable prestidigitation
Another possibility is that the NIC/driver and hub/switch may not auto-negotiate correctly. This may require that both ends are manually set for 100/Full to get the best and most reliable performance. Stan
On Monday 17 March 2003 09:51, Marco De Maddalena wrote:
I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with 2.4.x to copy a 36MB file - same HW) Does anybody have some hints, other than upgrade to newer kernel, since not always possible?
This is likely because the nfs server moved from user-space to kernel-space between 2.0.x and 2.4.x, ... Alternatively, are both machines actually doing exactly the same operation? Dylan -- "Every gun that is made, every warship launched, every rocket fired, signifies in a final sense a theft from those who hunger and are not fed—those who are cold and are not clothed. This world in arms is not spending its money alone—it is spending the sweat of its laborers, the genius of its scientists, the hopes of its children." —Dwight Eisenhower, Speech (1953)
On Monday 17 March 2003 03:51 am, Marco De Maddalena wrote:
I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with 2.4.x to copy a 36MB file - same HW) Does anybody have some hints, other than upgrade to newer kernel, since not always possible?
Thanks
I remember seeing some articles and research paper showing horrible 2.0.x kernel network performance especially when they are negotiating with windows hosts. Are the transfers between 2 Linux PC's or 2 Linux PC's with the same HW to a Windows Host. I'm not sure if 2.0.x /proc has the same setup as it does today, but try some parameter tuning in /proc/sys/net/ such as window size, TCPOPTs, etc. etc. There are many how-tos on what some of these ooption do. Maybe the 2.0.x kernel is using some funky or poorly tuned defaults? #------------------------ #Eric Bambach #Eric@CISU.net #------------------------
The 03.03.17 at 10:51, Marco De Maddalena wrote:
I have a server with SuSE8.1 I have noticed that Linux PCs with old kernel 2.0.x have bad network performance. (92 secs for a PC with 2.0.x against 7 secs for a PC with 2.4.x to copy a 36MB file - same HW)
Instead of transfering a file, where the throughput of the file system is involved, try "bing" instead, that tries to measure directly the network bandwidth. This is a trial run: cer@nimrodel:~> bing nimrodel telperion BING nimrodel.valinor (192.168.100.2) and telperion.valinor (192.168.100.1) 44 and 108 data bytes 1024 bits in 0.000ms 1024 bits in 0.005ms: 204800000bps, 0.000005ms per bit 1024 bits in 0.008ms: 128000000bps, 0.000008ms per bit 1024 bits in 0.017ms: 60235294bps, 0.000017ms per bit 1024 bits in 0.014ms: 73142857bps, 0.000014ms per bit 1024 bits in 0.015ms: 68266667bps, 0.000015ms per bit 1024 bits in 0.016ms: 64000000bps, 0.000016ms per bit ... [stop with ctrl-c] --- nimrodel.valinor statistics --- bytes out in dup loss rtt (ms): min avg max 44 19236 19236 0% 0.030 0.033 0.131 108 19236 19236 0% 0.027 0.029 3.860 --- telperion.valinor statistics --- bytes out in dup loss rtt (ms): min avg max 44 19236 19236 0% 0.211 0.262 71.571 108 19236 19235 0% 0.229 0.267 10.204 --- estimated link characteristics --- warning: rtt big host1 0.027ms < rtt small host2 0.030ms estimated throughput 56888889bps minimum delay per packet 0.169ms (9593 bits) average statistics (experimental) : packet loss: small 0%, big 0%, total 0% warning: rtt big host1 0.029ms < rtt small host2 0.033ms average throughput 204800000bps average delay per packet 0.217ms (12361 bits) weighted average throughput 204794677bps -- Cheers, Carlos Robinson
participants (7)
-
Carlos E. R. -
Dylan -
Eric -
Joris Verbogt -
Marco De Maddalena -
Simon Oliver -
Stan Glasoe