Date: Wed, 18 Sep 2002 21:32:08 -0700 From: Lars Eggert <larse@ISI.EDU> To: Steve Fettig <lists@stevenfettig.com> Cc: David Burns <david.burns@dugeem.net>, net@freebsd.org Subject: Re: Network Transfer Speed Issues - Tweaks/Advice? Message-ID: <3D895348.3000706@isi.edu> References: <F6AF9956-CB84-11D6-BDD9-00039384AB84@stevenfettig.com>
next in thread | previous in thread | raw e-mail | index | archive | help
This is a cryptographically signed message in MIME format. --------------ms050505020809000700020107 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Steve, your setup is way to complex. As David suggested, first try to establish whether the network connection between the two machines is the problem or not. Forget about NFS for now, just benchmark TCP and UDP (which do you NFS mount over, by the way?) If network performance is erratic, isolate the setup from cross traffic, and swap out NICs, hubs and or switches. Only after you've ruled out the network would I start worrying about disks, controllers and NFS settings. The idea here is to change ONE parameter in the setup at a time, and rerun your benchmark of choice. Steve Fettig wrote: > I really appreciate everyone who offered up a few ideas and answers. > One of the first things I finally figured out was that the adapter was > failing... hmmm doesn't make sense - it didn't fail completely, but I > swapped it out with a different adapter (same thing 2940UW) and the > performance change was quite remarkable. That still didn't solve the > huge discrepancy in speed difference between my other SCSI (internal) > drives and the ones I am working with, however. My next test will be to > build a system from the ground up w/ a 500 MHz processor and a 133 MHz > FSB. I am also wondering if I really missed the boat on the drives, > too. They are "older" 5400 RPM SCSI drives (HUGE drives physically) but > I don't see how they would be any slower than an IDE 5400 RPM drive - > which they are and quite a bit so. > As for all of the other stuff, the CPU load is usually at a steady 15% > (nfsd). Since that is the only service running (besides the required > daemons on a somewhat vanilla system) and the only load that the > "server" is under, I don't think that is an issue. I am going to test > system performance with the suggested programs (that was another thing I > was looking for in the answer, although I guess I didn't say it...) and > see what happens. I am starting to think that this is a combination of > controller and system not having the performance I was expecting. > Everything on the net end seems to be functioning without problem - no > lost packets, broken communication, etc. > > Well, the testing goes on. Please continue to comment if you have any > ideas (especially with regards to kernel tweaks, etc). > > Thanks, > Steve > > On Tuesday, September 17, 2002, at 07:55 , David Burns wrote: > >> Steve Fettig wrote: >> >>> I recently set up an NFS server to run daily backups on. The server >>> was built using an old P150 w/ 90 MB of ram and a 6GB hard drive. >>> (All servers in this experiment are set up using FBSD 4.6.2 and the >>> client is a Mac PowerBook G4 running Mac OS X.) Attached to it is an >>> external SCSI hard drive enclosure with 4 47GB SCSI drives running >>> off an AHA-2490UW SCSI adapter. I am getting really odd performance >>> when doing an NFS transfer (I also get odd performance out of scp) >>> from the machine I am trying to back up. I will get a burst of >>> 20Mbps for about 30 seconds, then it will ramp down to 1 Mbps for >>> about 2 minutes, ramp backup to 20 Mbps, then back down to 1 Mbps and >>> so on. >> >> You need to break the problem down ... Is the system CPU and/or IO >> bound during the backup? Also try some quick benchmarks to verify >> basic system performance levels: Network IO - use ttcp (or netperf >> etc), and Disk IO - use bonnie (or similar). >> >> NB Of course you can't simply take such benchmarks results and put >> them together - but you will gain a better understanding of where the >> potential slowdowns may be. >> >> Lastly, performance issues on older Pentiums can also result from poor >> memory bandwidth and/or PCI chipset problems. I recently replaced a >> P120 with a Celeron 333 - the performance improvement was surprising. >> >> Regards, >> David >> > > > To Unsubscribe: send mail to majordomo@FreeBSD.org > with "unsubscribe freebsd-net" in the body of the message -- Lars Eggert <larse@isi.edu> USC Information Sciences Institute --------------ms050505020809000700020107 Content-Type: application/x-pkcs7-signature; name="smime.p7s" Content-Transfer-Encoding: base64 Content-Disposition: attachment; filename="smime.p7s" Content-Description: S/MIME Cryptographic Signature MIAGCSqGSIb3DQEHAqCAMIACAQExCzAJBgUrDgMCGgUAMIAGCSqGSIb3DQEHAQAAoIIJtjCC AzgwggKhoAMCAQICEGZFcrfMdPXPY3ZFhNAukQEwDQYJKoZIhvcNAQEEBQAwgdExCzAJBgNV BAYTAlpBMRUwEwYDVQQIEwxXZXN0ZXJuIENhcGUxEjAQBgNVBAcTCUNhcGUgVG93bjEaMBgG A1UEChMRVGhhd3RlIENvbnN1bHRpbmcxKDAmBgNVBAsTH0NlcnRpZmljYXRpb24gU2Vydmlj ZXMgRGl2aXNpb24xJDAiBgNVBAMTG1RoYXd0ZSBQZXJzb25hbCBGcmVlbWFpbCBDQTErMCkG CSqGSIb3DQEJARYccGVyc29uYWwtZnJlZW1haWxAdGhhd3RlLmNvbTAeFw0wMDA4MzAwMDAw MDBaFw0wNDA4MjcyMzU5NTlaMIGSMQswCQYDVQQGEwJaQTEVMBMGA1UECBMMV2VzdGVybiBD YXBlMRIwEAYDVQQHEwlDYXBlIFRvd24xDzANBgNVBAoTBlRoYXd0ZTEdMBsGA1UECxMUQ2Vy dGlmaWNhdGUgU2VydmljZXMxKDAmBgNVBAMTH1BlcnNvbmFsIEZyZWVtYWlsIFJTQSAyMDAw LjguMzAwgZ8wDQYJKoZIhvcNAQEBBQADgY0AMIGJAoGBAN4zMqZjxwklRT7SbngnZ4HF2ogZ gpcO40QpimM1Km1wPPrcrvfudG8wvDOQf/k0caCjbZjxw0+iZdsN+kvx1t1hpfmFzVWaNRqd knWoJ67Ycvm6AvbXsJHeHOmr4BgDqHxDQlBRh4M88Dm0m1SKE4f/s5udSWYALQmJ7JRr6aFp AgMBAAGjTjBMMCkGA1UdEQQiMCCkHjAcMRowGAYDVQQDExFQcml2YXRlTGFiZWwxLTI5NzAS BgNVHRMBAf8ECDAGAQH/AgEAMAsGA1UdDwQEAwIBBjANBgkqhkiG9w0BAQQFAAOBgQAxsUtH XfkBceX1U2xdedY9mMAmE2KBIqcS+CKV6BtJtyd7BDm6/ObyJOuR+r3sDSo491BVqGz3Da1M G7wD9LXrokefbKIMWI0xQgkRbLAaadErErJAXWr5edDqLiXdiuT82w0fnQLzWtvKPPZE6iZp h39Ins6ln+eE2MliYq0FxjCCAzkwggKioAMCAQICAwglQTANBgkqhkiG9w0BAQQFADCBkjEL MAkGA1UEBhMCWkExFTATBgNVBAgTDFdlc3Rlcm4gQ2FwZTESMBAGA1UEBxMJQ2FwZSBUb3du MQ8wDQYDVQQKEwZUaGF3dGUxHTAbBgNVBAsTFENlcnRpZmljYXRlIFNlcnZpY2VzMSgwJgYD VQQDEx9QZXJzb25hbCBGcmVlbWFpbCBSU0EgMjAwMC44LjMwMB4XDTAyMDgyNDE4NTMzOVoX DTAzMDgyNDE4NTMzOVowVDEPMA0GA1UEBBMGRWdnZXJ0MQ0wCwYDVQQqEwRMYXJzMRQwEgYD VQQDEwtMYXJzIEVnZ2VydDEcMBoGCSqGSIb3DQEJARYNbGFyc2VAaXNpLmVkdTCCASIwDQYJ KoZIhvcNAQEBBQADggEPADCCAQoCggEBANI2Rrt4ggaQ/IrOsDeOm2H4/R5FRIL6JjDY3StE aogp1r23WKniQ1Vj98Nu5WxlaZ3Iam3Jen5T66H8u7rtMNpK4qAeAGoBsVeyVr1+CTFeuv+m xCh7BvBJwhLdm0zDaoDT05YKYZaqtsT+F286FWJQg31Xtf+vTKLVVrHcsafnteyal2NEt7Ac yZZfjsVLwxp2Lq3cwYfRQRoo7/yCVzS7HsgM6jmbO4taEMo4yC2rpnUbWEUCDTaCYgpAXzAl oiNk7GDh0wz2s5ZSnHRvNSBMAjCmpNtSYHfXFI1ANwrrrHIJ7Ei83+XN32PWY4OPzO3iown9 VR+vM+8lNx9OX28CAwEAAaNWMFQwKgYFK2UBBAEEITAfAgEAMBowGAIBBAQTTDJ1TXlmZkJO VWJOSkpjZFoyczAYBgNVHREEETAPgQ1sYXJzZUBpc2kuZWR1MAwGA1UdEwEB/wQCMAAwDQYJ KoZIhvcNAQEEBQADgYEAXcrIlKmPLM/r8r3oz2ZLPLaT1AyMjYTZY2qq/R7SUtFa9BNlTIFh DG78QKfJ9lo2LMzTPQqMZgNLmj95GbNPI8P8OIq2K6MeCZWz08ROackqTFP6xWbIFIfXcBVR 1dZnDDyDKBBh05KkvyTPawSQyOBUeNBfQUyO4TE+3o58U8UwggM5MIICoqADAgECAgMIJUEw DQYJKoZIhvcNAQEEBQAwgZIxCzAJBgNVBAYTAlpBMRUwEwYDVQQIEwxXZXN0ZXJuIENhcGUx EjAQBgNVBAcTCUNhcGUgVG93bjEPMA0GA1UEChMGVGhhd3RlMR0wGwYDVQQLExRDZXJ0aWZp Y2F0ZSBTZXJ2aWNlczEoMCYGA1UEAxMfUGVyc29uYWwgRnJlZW1haWwgUlNBIDIwMDAuOC4z MDAeFw0wMjA4MjQxODUzMzlaFw0wMzA4MjQxODUzMzlaMFQxDzANBgNVBAQTBkVnZ2VydDEN MAsGA1UEKhMETGFyczEUMBIGA1UEAxMLTGFycyBFZ2dlcnQxHDAaBgkqhkiG9w0BCQEWDWxh cnNlQGlzaS5lZHUwggEiMA0GCSqGSIb3DQEBAQUAA4IBDwAwggEKAoIBAQDSNka7eIIGkPyK zrA3jpth+P0eRUSC+iYw2N0rRGqIKda9t1ip4kNVY/fDbuVsZWmdyGptyXp+U+uh/Lu67TDa SuKgHgBqAbFXsla9fgkxXrr/psQoewbwScIS3ZtMw2qA09OWCmGWqrbE/hdvOhViUIN9V7X/ r0yi1Vax3LGn57XsmpdjRLewHMmWX47FS8Madi6t3MGH0UEaKO/8glc0ux7IDOo5mzuLWhDK OMgtq6Z1G1hFAg02gmIKQF8wJaIjZOxg4dMM9rOWUpx0bzUgTAIwpqTbUmB31xSNQDcK66xy CexIvN/lzd9j1mODj8zt4qMJ/VUfrzPvJTcfTl9vAgMBAAGjVjBUMCoGBStlAQQBBCEwHwIB ADAaMBgCAQQEE0wydU15ZmZCTlViTkpKY2RaMnMwGAYDVR0RBBEwD4ENbGFyc2VAaXNpLmVk dTAMBgNVHRMBAf8EAjAAMA0GCSqGSIb3DQEBBAUAA4GBAF3KyJSpjyzP6/K96M9mSzy2k9QM jI2E2WNqqv0e0lLRWvQTZUyBYQxu/ECnyfZaNizM0z0KjGYDS5o/eRmzTyPD/DiKtiujHgmV s9PETmnJKkxT+sVmyBSH13AVUdXWZww8gygQYdOSpL8kz2sEkMjgVHjQX0FMjuExPt6OfFPF MYIDJzCCAyMCAQEwgZowgZIxCzAJBgNVBAYTAlpBMRUwEwYDVQQIEwxXZXN0ZXJuIENhcGUx EjAQBgNVBAcTCUNhcGUgVG93bjEPMA0GA1UEChMGVGhhd3RlMR0wGwYDVQQLExRDZXJ0aWZp Y2F0ZSBTZXJ2aWNlczEoMCYGA1UEAxMfUGVyc29uYWwgRnJlZW1haWwgUlNBIDIwMDAuOC4z MAIDCCVBMAkGBSsOAwIaBQCgggFhMBgGCSqGSIb3DQEJAzELBgkqhkiG9w0BBwEwHAYJKoZI hvcNAQkFMQ8XDTAyMDkxOTA0MzIwOFowIwYJKoZIhvcNAQkEMRYEFFBcbECimnFwfNPqmFAH k+YyVSNOMFIGCSqGSIb3DQEJDzFFMEMwCgYIKoZIhvcNAwcwDgYIKoZIhvcNAwICAgCAMA0G CCqGSIb3DQMCAgFAMAcGBSsOAwIHMA0GCCqGSIb3DQMCAgEoMIGtBgsqhkiG9w0BCRACCzGB naCBmjCBkjELMAkGA1UEBhMCWkExFTATBgNVBAgTDFdlc3Rlcm4gQ2FwZTESMBAGA1UEBxMJ Q2FwZSBUb3duMQ8wDQYDVQQKEwZUaGF3dGUxHTAbBgNVBAsTFENlcnRpZmljYXRlIFNlcnZp Y2VzMSgwJgYDVQQDEx9QZXJzb25hbCBGcmVlbWFpbCBSU0EgMjAwMC44LjMwAgMIJUEwDQYJ KoZIhvcNAQEBBQAEggEAE2zZ20onfGzhWgb9bpf8WZ2Bph4zLInjaJj4iP3nyig7OIDWF0Qh WzvqszYB0BcVIJE2TuEjyBCfDW0J1XnELIgin7WxFqlnhhGAhA3IBApj6y8HIQaBuCDGbpvb 7h0ojnRfdoe2uCH5hqinv6Pmz2VmO1CLJ02UtO5LJqQWy4bSYAlhAaiik77d9KD88k4uGufQ itGoW2Tu3QjBJTvhC2TyqqjAGGZAVgLW6aDnUQqX0+nTCoqpxNwhCDJcq/Wgs9LEsbEctwFK bYAbOyeasXATjHBi7nIm7wWHEjvJyzmyUnqC/jb94pj5tIWmjMC8c8trwuXwmNVq77UFo0hf zwAAAAAAAA== --------------ms050505020809000700020107-- To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3D895348.3000706>