Re: NFS performances on 5.1
On a same Ethernet 100 LAN, I have several Unix and an NFS Solaris fileserver. On these Unix, I tried: time dd=/fileserver/aFile of=/fileserver/otherFile bs=32768 On each try, I use new files, to avoid the impact of file caching. I measured the time spent and the number of Ethernet packets (with snoop). I found: NFS client time# pkts === === == Solaris3.11s 2296 Linux Redhat9 2.42s 1929 FreeBSD 5.119.72s 14887 !!! FreeBSD 4.93.04s 6380 FreeBSD 5.22.98s 5941 The best way to tune 5.1 is to update it to 5.2 (I'm sure you read all the documentation that states that the 5.x branch is a new technology release with performance not being an initial goal). Of course, but I have 60+ stations tu upgrade... However, those numbers still look excessive, so I wonder if you forgot to turn off some of the debugging options like WITNESS. I have the kernel from the distribution. -- Jacques Beigbeder| [EMAIL PROTECTED] Service de Prestations Informatiques | http://www.spi.ens.fr Ecole normale supérieure | 45 rue d'Ulm |Tel : (+33 1)1 44 32 37 96 F75230 Paris cedex 05|Fax : (+33 1)1 44 32 20 75 ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
NFS performances on 5.1
On a same Ethernet 100 LAN, I have several Unix and an NFS Solaris fileserver. On these Unix, I tried: time dd=/fileserver/aFile of=/fileserver/otherFile bs=32768 On each try, I use new files, to avoid the impact of file caching. I measured the time spent and the number of Ethernet packets (with snoop). I found: NFS client time# pkts === === == Solaris 3.11s 2296 Linux Redhat9 2.42s 1929 FreeBSD 5.1 19.72s 14887 !!! FreeBSD 4.9 3.04s 6380 FreeBSD 5.2 2.98s 5941 All FreeBSD uses: mount_nfs -U -3 -r 32768 -w 32768 ... Question: is there any tuning on 5.1 to get better performances? -- Jacques Beigbeder| [EMAIL PROTECTED] Service de Prestations Informatiques | http://www.spi.ens.fr Ecole normale supérieure | 45 rue d'Ulm |Tel : (+33 1)1 44 32 37 96 F75230 Paris cedex 05|Fax : (+33 1)1 44 32 20 75 ___ [EMAIL PROTECTED] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-questions To unsubscribe, send any mail to [EMAIL PROTECTED]
Re: NFS performances on 5.1
On Saturday 31 January 2004 16:02, Jacques Beigbeder wrote: time dd=/fileserver/aFile of=/fileserver/otherFile bs=32768 NFS clienttime# pkts === === == Solaris 3.11s 2296 Linux Redhat9 2.42s 1929 FreeBSD 5.1 19.72s 14887 !!! FreeBSD 4.9 3.04s 6380 FreeBSD 5.2 2.98s 5941 All FreeBSD uses: mount_nfs -U -3 -r 32768 -w 32768 ... Question: is there any tuning on 5.1 to get better performances? Did you read the notes in src/UPDATING saying: NOTE TO PEOPLE WHO THINK THAT FreeBSD 5.x IS SLOW: -- Melvyn === FreeBSD sarevok.webteckies.org 5.2-CURRENT FreeBSD 5.2-CURRENT #0: Wed Jan 28 18:01:18 CET 2004 [EMAIL PROTECTED]:/usr/obj/usr/src/sys/SAREVOK_NOAPM_NODEBUG i386 === pgp0.pgp Description: signature
Re: NFS performances on 5.1
On Sat, Jan 31, 2004 at 04:02:21PM +0100, Jacques Beigbeder wrote: On a same Ethernet 100 LAN, I have several Unix and an NFS Solaris fileserver. On these Unix, I tried: time dd=/fileserver/aFile of=/fileserver/otherFile bs=32768 On each try, I use new files, to avoid the impact of file caching. I measured the time spent and the number of Ethernet packets (with snoop). I found: NFS clienttime# pkts === === == Solaris 3.11s 2296 Linux Redhat9 2.42s 1929 FreeBSD 5.1 19.72s 14887 !!! FreeBSD 4.9 3.04s 6380 FreeBSD 5.2 2.98s 5941 All FreeBSD uses: mount_nfs -U -3 -r 32768 -w 32768 ... Question: is there any tuning on 5.1 to get better performances? The best way to tune 5.1 is to update it to 5.2 (I'm sure you read all the documentation that states that the 5.x branch is a new technology release with performance not being an initial goal). However, those numbers still look excessive, so I wonder if you forgot to turn off some of the debugging options like WITNESS. Kris pgp0.pgp Description: PGP signature
Re: NFS performances on 5.1
On Saturday 31 January 2004 17:11, Melvyn Sopacua wrote: On Saturday 31 January 2004 16:02, Jacques Beigbeder wrote: time dd=/fileserver/aFile of=/fileserver/otherFile bs=32768 NFS client time# pkts === === == Solaris 3.11s 2296 Linux Redhat9 2.42s 1929 FreeBSD 5.1 19.72s 14887 !!! FreeBSD 4.9 3.04s 6380 FreeBSD 5.2 2.98s 5941 All FreeBSD uses: mount_nfs -U -3 -r 32768 -w 32768 ... Question: is there any tuning on 5.1 to get better performances? Did you read the notes in src/UPDATING saying: NOTE TO PEOPLE WHO THINK THAT FreeBSD 5.x IS SLOW: Could you pleas explain that numbers? I did almost the same test and found the following values in MByte/s: FBSD 5.2 - 5.2 / 4,6(Client 1,1G Cel, Server C3 800) Linux- 5.2 / lockup (Client 733 PIII, Server C3 800) FBSD 4.9 - 5.2 / 2,8(CLient 233MMX, Server C3 800) FBSD 5.2 - 5.1 / 6,5(Client 1,1G Cel, Server C3 800) Linux- 5.1 / 9,8(Client 733 PIII, Server C3 800) FBSD 5.2 - 4.9 / 5,8(Client 1,1G Cel, Server C3 800) Linux- 4.9 / 9,8(Client 733 PIII, Server C3 800) FBSD 4.9 - 4.9 / 3,0(Client 233MMX, Server C3 800) DragonFlyBSD as Server resulted in about 5% more performance than 4.9 (linear exept Linux Client as it performs with the maximum Ethernet Speed) My tests were without modifying any rsize/wsize. But even with (rw)size 32k i had expected to be able to transfer about 9 MByte/s from a 233MMX box. 3MByte/s is absolutely lousy. What hardware do we need for just tranfsering bytes? 1GHz? I think 233 MHz with 64MB for OS should be more than enough. Regrettably I haven't had time to install a Linux on the 233 box. Btw. Linux = DebWoody (2.4.22) and all clients have fxp interfaces! Summary: 5.1 as server was a lot faster than 5.2 as server so is 4.9. Fastest was DragonFlyBSD but anyhow, just Linux as Client does a reasonable job. And not to forget the broken Linux - 5.2 support!!! -Harry pgp0.pgp Description: signature
Re: NFS performances on 5.1
On Saturday 31 January 2004 19:03, Harald Schmalzbauer wrote: On Saturday 31 January 2004 17:11, Melvyn Sopacua wrote: Did you read the notes in src/UPDATING saying: NOTE TO PEOPLE WHO THINK THAT FreeBSD 5.x IS SLOW: Could you pleas explain that numbers? No, because you also failed to mention if you've read that section and disabled WITNESS and so on. -- Melvyn === FreeBSD sarevok.webteckies.org 5.2-CURRENT FreeBSD 5.2-CURRENT #0: Wed Jan 28 18:01:18 CET 2004 [EMAIL PROTECTED]:/usr/obj/usr/src/sys/SAREVOK_NOAPM_NODEBUG i386 === pgp0.pgp Description: signature
Re: NFS performances on 5.1
On Sunday 01 February 2004 00:26, Melvyn Sopacua wrote: On Saturday 31 January 2004 19:03, Harald Schmalzbauer wrote: On Saturday 31 January 2004 17:11, Melvyn Sopacua wrote: Did you read the notes in src/UPDATING saying: NOTE TO PEOPLE WHO THINK THAT FreeBSD 5.x IS SLOW: Could you pleas explain that numbers? No, because you also failed to mention if you've read that section and disabled WITNESS and so on. Ok, it was another thread where I mentioned that I'm tracking current for a long time, but of course I read it, I even can't remember when it was defaulted to be excluded. So I have code from today and I use similar generic options for today in my kernel. Btw, I replyed to the wrong posting, I asked Jacques Beigbeder to explain his numbers. And even with WITNESS and INVARIANTS in the kernel it would be poor performance since the difference is just about 10-20 percent. (last time I checked, which is a while ago) -Harry pgp0.pgp Description: signature
Re: NFS performances on 5.1
On Saturday 31 January 2004 19:03, Harald Schmalzbauer wrote: On Saturday 31 January 2004 17:11, Melvyn Sopacua wrote: On Saturday 31 January 2004 16:02, Jacques Beigbeder wrote: time dd=/fileserver/aFile of=/fileserver/otherFile bs=32768 NFS clienttime# pkts === === == Solaris 3.11s 2296 Linux Redhat9 2.42s 1929 FreeBSD 5.1 19.72s 14887 !!! FreeBSD 4.9 3.04s 6380 FreeBSD 5.2 2.98s 5941 All FreeBSD uses: mount_nfs -U -3 -r 32768 -w 32768 ... Question: is there any tuning on 5.1 to get better performances? Did you read the notes in src/UPDATING saying: NOTE TO PEOPLE WHO THINK THAT FreeBSD 5.x IS SLOW: Could you pleas explain that numbers? I did almost the same test and found the following values in MByte/s: FBSD 5.2 - 5.2 / 4,6(Client 1,1G Cel, Server C3 800) Linux- 5.2 / lockup (Client 733 PIII, Server C3 800) FBSD 4.9 - 5.2 / 2,8(CLient 233MMX, Server C3 800) FBSD 5.2 - 5.1 / 6,5(Client 1,1G Cel, Server C3 800) Linux- 5.1 / 9,8(Client 733 PIII, Server C3 800) FBSD 5.2 - 4.9 / 5,8(Client 1,1G Cel, Server C3 800) Linux- 4.9 / 9,8(Client 733 PIII, Server C3 800) FBSD 4.9 - 4.9 / 3,0(Client 233MMX, Server C3 800) DragonFlyBSD as Server resulted in about 5% more performance than 4.9 (linear exept Linux Client as it performs with the maximum Ethernet Speed) My tests were without modifying any rsize/wsize. But even with (rw)size 32k i had expected to be able to transfer about 9 MByte/s from a 233MMX box. 3MByte/s is absolutely lousy. What hardware do we need for just tranfsering bytes? 1GHz? I think 233 MHz with 64MB for OS should be more than enough. Regrettably I haven't had time to install a Linux on the 233 box. Btw. Linux = DebWoody (2.4.22) and all clients have fxp interfaces! Summary: 5.1 as server was a lot faster than 5.2 as server so is 4.9. Fastest was DragonFlyBSD but anyhow, just Linux as Client does a reasonable job. And not to forget the broken Linux - 5.2 support!!! Oh, I forgot to mention one very important thing: The idle cycles while NFS transfers. Like mentioned, the server is a C3 800 Processor with 256MB RAM nothing doing else than feeding NFS and SMB Clients (via SAMBA 3.0.1). In case of my NFS mesurements, 5.2 had 20% idle while feeding 5.2 with 3,5MByte/s!!! Best was Dragonfly which had 60% idle when feeding the Linux box with 9.8MByte/s while 4.9 only had 50% idle (while feeding the Linux Client with 9.8Mbyte/s). -Harry pgp0.pgp Description: signature