Re: mbuf leak with rl
On Wednesday 20 September 2006 20:36, Matthew R. Dempsky wrote: > On Wed, Sep 20, 2006 at 10:29:10AM -0500, Karle, Chris wrote: > > That looks suspect to me; that seems like a lot for cable modem level > > traffic. > > > > I'd check if your mbufs number ever goes down. > > I've rechecked the output of netstat -m occasionally since then, and I > haven't seen them go down at all--only steadily increase. As of > typing this email, the output is: > > $ netstat -m > 3616 mbufs in use: > 3593 mbufs allocated to data > 6 mbufs allocated to packet headers > 17 mbufs allocated to socket names and addresses > 855/870/6144 mbuf clusters in use (current/peak/max) > 2656 Kbytes allocated to network (98% in use) > 0 requests for memory denied > 0 requests for memory delayed > 0 calls to protocol drain routines Same story, rl on cable modem, I do see it oscillating a bit, but the tendency is steadily up: 1834 mbufs in use: 1655 mbufs allocated to data 14 mbufs allocated to packet headers 165 mbufs allocated to socket names and addresses 428/658/6144 mbuf clusters in use (current/peak/max) 1812 Kbytes allocated to network (72% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines Compared to 1500 from a week ago. (no reboot in between) -- viq
Re: mbuf leak with rl
Looks like you're experiencing the same quirk that me and another gentlemen have. We all have rl interfaces on cable modems. I replaced my rl with a different interface and have had no problems since. -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Matthew R. Dempsky Sent: Wednesday, September 20, 2006 1:37 PM To: misc@openbsd.org Subject: Re: mbuf leak with rl On Wed, Sep 20, 2006 at 10:29:10AM -0500, Karle, Chris wrote: > That looks suspect to me; that seems like a lot for cable modem level > traffic. > > I'd check if your mbufs number ever goes down. I've rechecked the output of netstat -m occasionally since then, and I haven't seen them go down at all--only steadily increase. As of typing this email, the output is: $ netstat -m 3616 mbufs in use: 3593 mbufs allocated to data 6 mbufs allocated to packet headers 17 mbufs allocated to socket names and addresses 855/870/6144 mbuf clusters in use (current/peak/max) 2656 Kbytes allocated to network (98% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines
Re: mbuf leak with rl
On Wed, Sep 20, 2006 at 10:29:10AM -0500, Karle, Chris wrote: > That looks suspect to me; that seems like a lot for cable modem level > traffic. > > I'd check if your mbufs number ever goes down. I've rechecked the output of netstat -m occasionally since then, and I haven't seen them go down at all--only steadily increase. As of typing this email, the output is: $ netstat -m 3616 mbufs in use: 3593 mbufs allocated to data 6 mbufs allocated to packet headers 17 mbufs allocated to socket names and addresses 855/870/6144 mbuf clusters in use (current/peak/max) 2656 Kbytes allocated to network (98% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines
Re: mbuf leak with rl
That looks suspect to me; that seems like a lot for cable modem level traffic. I'd check if your mbufs number ever goes down. -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Matthew R. Dempsky Sent: Friday, September 15, 2006 4:46 PM To: misc@openbsd.org Subject: Re: mbuf leak with rl On Thu, Sep 14, 2006 at 10:38:35AM -0500, Karle, Chris wrote: > If you're using a "rl*" can you take a look at your mbuf usage (netstat -m)? On my OpenBSD 3.9 firewall, sis0 is connected to my internal network, and rl0 is connected to my cable modem. $ netstat -m 2546 mbufs in use: 2525 mbufs allocated to data 5 mbufs allocated to packet headers 16 mbufs allocated to socket names and addresses 630/648/6144 mbuf clusters in use (current/peak/max) 1952 Kbytes allocated to network (97% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines $ dmesg | grep -e GENERIC -e rl -e sis OpenBSD 3.9 (GENERIC) #617: Thu Mar 2 02:26:48 MST 2006 [EMAIL PROTECTED]:/usr/src/sys/arch/i386/compile/GENERIC sis0 at pci0 dev 4 function 0 "SiS 900 10/100BaseTX" rev 0x91: irq 11, address 00:14:2a:b7:c9:17 rlphy0 at sis0 phy 9: RTL8201L 10/100 PHY, rev. 1 rl0 at pci0 dev 11 function 0 "Accton MPX 5030/5038" rev 0x10: irq 11, address 00:e0:29:58:9b:eb rlphy1 at rl0 phy 0: RTL internal PHY
Re: mbuf leak with rl
On 2006/09/14 10:38, Karle, Chris wrote: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that mbufs > will slowly leak when using it. Not a fix for rl, but if it's a newer one (8139C+ or 8101E) you should find that upgrading to a snapshot changes it to re(4), which uses a different method of accessing the card with better performance (e.g. better DMA, checksum offload, hw vlan tagging). Might be timely to point this out now, in case anyone does a remote upgrade on a box with one of these and ends up with no network because they didn't copy hostname.rl0 to hostname.re0 before they reboot with the new kernel.
Re: mbuf leak with rl
On Thu, Sep 14, 2006 at 10:38:35AM -0500, Karle, Chris wrote: > If you're using a "rl*" can you take a look at your mbuf usage (netstat -m)? On my OpenBSD 3.9 firewall, sis0 is connected to my internal network, and rl0 is connected to my cable modem. $ netstat -m 2546 mbufs in use: 2525 mbufs allocated to data 5 mbufs allocated to packet headers 16 mbufs allocated to socket names and addresses 630/648/6144 mbuf clusters in use (current/peak/max) 1952 Kbytes allocated to network (97% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines $ dmesg | grep -e GENERIC -e rl -e sis OpenBSD 3.9 (GENERIC) #617: Thu Mar 2 02:26:48 MST 2006 [EMAIL PROTECTED]:/usr/src/sys/arch/i386/compile/GENERIC sis0 at pci0 dev 4 function 0 "SiS 900 10/100BaseTX" rev 0x91: irq 11, address 00:14:2a:b7:c9:17 rlphy0 at sis0 phy 9: RTL8201L 10/100 PHY, rev. 1 rl0 at pci0 dev 11 function 0 "Accton MPX 5030/5038" rev 0x10: irq 11, address 00:e0:29:58:9b:eb rlphy1 at rl0 phy 0: RTL internal PHY
Re: mbuf leak with rl
I posted about my mbuf leak problem earlier, but I thought I'd chime in again. For those without uptimes, the mbuf usage depends on the uptime of the system and is pretty meaningless if you just restarted. Uptime: 1:30AM up 8 days, 5:12, 0 users, load averages: 0.13, 0.11, 0.08 Netstat -m: 12549 mbufs in use: 12545 mbufs allocated to data 4 mbufs allocated to socket names and addresses 2706/2716/15000 mbuf clusters in use (current/peak/max) 8576 Kbytes allocated to network (99% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines dmesg | grep rl: rl0 at pci1 dev 13 function 0 "Realtek 8139" rev 0x10: irq 9, address 00:e0:4c:e4:f1:5d rlphy0 at rl0 phy 0: RTL internal PHY rl1 at pci1 dev 14 function 0 "D-Link Systems 530TX+" rev 0x10: irq 3, address 00:50:ba:55:10:f1 rlphy1 at rl1 phy 0: RTL internal PHY Though, I don't get such horrible performance consistently (but there's always a leak). Note that in thise case kern.maxclusters was increased to 15000 for testing purposes. -Stefan
Re: mbuf leak with rl
On Thu, Sep 14, 2006 at 10:38:35AM -0500, Karle, Chris wrote: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that mbufs > will slowly leak when using it. I noticed this after switching to 3.9. I I have 2 rl and 1 sk interface in my AMD64 machine, and this works fine. Home usage, so no heavy traffic, but uptimes of 50-60 days with no problems. $ dmesg | grep rl ; uptime ; netstat -m rl0 at pci0 dev 12 function 0 "Realtek 8139" rev 0x10: irq 10, address 00:00:b4:93:54:c4 rlphy0 at rl0 phy 0: RTL internal PHY rl1 at pci0 dev 13 function 0 "Realtek 8139" rev 0x10: irq 5, address 00:e0:4c:49:78:1d rlphy1 at rl1 phy 0: RTL internal PHY 7:59PM up 7 days, 7:22, 1 user, load averages: 0.45, 0.50, 0.47 264 mbufs in use: 256 mbufs allocated to data 2 mbufs allocated to packet headers 6 mbufs allocated to socket names and addresses 0/120/6144 mbuf clusters in use (current/peak/max) 344 Kbytes allocated to network (19% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines -- Jurjen Oskam Savage's Law of Expediency: You want it bad, you'll get it bad.
Re: mbuf leak with rl
I forgot to mention that my rl interface is on a cable modem, which tends to have a lot of ARP traffic. -Original Message- From: Abel Talaversn Estevez [mailto:[EMAIL PROTECTED] Sent: Thursday, September 14, 2006 11:14 AM To: Karle, Chris Cc: misc@openbsd.org Subject: Re: mbuf leak with rl El Jueves, 14 de Septiembre de 2006 17:38, escribis: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that > mbufs will slowly leak when using it. I noticed this after switching > to 3.9. I don't know if something happened to the card or not... > maybe there is a hardware error now that is making it behave funky. > > If you're using a "rl*" can you take a look at your mbuf usage > (netstat -m)? Me and another person both see something similar. > > Thanks, > Chris > > > dmesg: > rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 11, address > 00:48:54:65:39:5a rlphy0 at rl0 phy 0: RTL internal PHY Look, I have a realtek NIC in OpenBSD 3.7 and OpenBSD 3.9: OpenBSD 3.9: # dmesg | grep rl rl0 at pci0 dev 13 function 0 "D-Link Systems 530TX+" rev 0x10: irq 11, address 00:0d:88:1a:8e:3a rlphy0 at rl0 phy 0: RTL internal PHY rlphy1 at vr0 phy 1: RTL8201L 10/100 PHY, rev. 1 # netstat -m 4 mbufs in use: 1 mbuf allocated to packet headers 3 mbufs allocated to socket names and addresses 0/10/6144 mbuf clusters in use (current/peak/max) 28 Kbytes allocated to network (3% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines # OpenBSD 3.7: # netstat -m 12 mbufs in use: 1 mbuf allocated to packet headers 11 mbufs allocated to socket names and addresses 0/64/6144 mbuf clusters in use (current/peak/max) 168 Kbytes allocated to network (1% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines # dmesg | grep rl rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 12 address 00:03:2d:04:60:40 rlphy0 at rl0 phy 0: RTL internal phy rl1 at pci0 dev 9 function 0 "Realtek 8139" rev 0x10: irq 10 address 00:03:2d:04:60:3f rlphy1 at rl1 phy 0: RTL internal phy rl2 at pci0 dev 10 function 0 "Realtek 8139" rev 0x10: irq 11 address 00:03:2d:04:60:3e rlphy2 at rl2 phy 0: RTL internal phy rl3 at pci0 dev 11 function 0 "Realtek 8139" rev 0x10: irq 15 address 00:03:2d:04:60:3d rlphy3 at rl3 phy 0: RTL internal phy What do you think? -- Abel Talaversn Estevez Ingeniero Superior de Telecomunicaciones Analista de Proyectos OpenWired Caballero 87 - Bajos 08029 - Barcelona Tel. 93 495 0990 Fax. 93 419 4591 http://www.openwired.com
Re: mbuf leak with rl
I mentioned this in a different post too; I should have included it in my original message. My rl interface is on a cable modem, which tend to be very chatty with ARP traffic. The output of netstat -m ever increases; I ran a cronjob which captured it. After about 10-12 days the network would die because of lack of mbufs. This morning I switched to a "dc" interface and my mbufs usage has not moved from 135. Thanks, Chris -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Hans van Leeuwen Sent: Thursday, September 14, 2006 11:27 AM To: misc@openbsd.org Subject: Re: mbuf leak with rl On Thursday 14 September 2006 17:38, you wrote: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that > mbufs will slowly leak when using it. I noticed this after switching > to 3.9. I don't know if something happened to the card or not... > maybe there is a hardware error now that is making it behave funky. > > If you're using a "rl*" can you take a look at your mbuf usage > (netstat -m)? Me and another person both see something similar. 237 mbufs in use: 135 mbufs allocated to data 66 mbufs allocated to packet headers 36 mbufs allocated to socket names and addresses 125/380/6144 mbuf clusters in use (current/peak/max) 856 Kbytes allocated to network (36% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines I have no idea if this is good or not. > dmesg: > rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 11, address > 00:48:54:65:39:5a rlphy0 at rl0 phy 0: RTL internal PHY rl0 at pci1 dev 10 function 0 "Realtek 8139" rev 0x10: irq 11, address 00:10:a7:0b:16:ed rlphy0 at rl0 phy 0: RTL internal PHY Greetings, Hans
Re: mbuf leak with rl
On Thu, Sep 14, 2006 at 10:38:35AM -0500, Karle, Chris wrote: > If you're using a "rl*" can you take a look at your mbuf usage (netstat -m)? > Me and another person both see something similar. OpenBSD 3.9-stable (i386 GENERIC) % dmesg | grep rl rl0 at pci0 dev 14 function 0 "Realtek 8139" rev 0x10: irq 11, address 00:30:1b:0f:e1:aa rlphy0 at rl0 phy 0: RTL internal PHY % netstat -m 75 mbufs in use: 69 mbufs allocated to data 2 mbufs allocated to packet headers 4 mbufs allocated to socket names and addresses 68/146/6144 mbuf clusters in use (current/peak/max) 344 Kbytes allocated to network (44% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines
Re: mbuf leak with rl
On Thursday 14 September 2006 17:38, you wrote: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that mbufs > will slowly leak when using it. I noticed this after switching to 3.9. I > don't know if something happened to the card or not... maybe there is a > hardware error now that is making it behave funky. > > If you're using a "rl*" can you take a look at your mbuf usage (netstat > -m)? Me and another person both see something similar. 237 mbufs in use: 135 mbufs allocated to data 66 mbufs allocated to packet headers 36 mbufs allocated to socket names and addresses 125/380/6144 mbuf clusters in use (current/peak/max) 856 Kbytes allocated to network (36% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines I have no idea if this is good or not. > dmesg: > rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 11, address > 00:48:54:65:39:5a > rlphy0 at rl0 phy 0: RTL internal PHY rl0 at pci1 dev 10 function 0 "Realtek 8139" rev 0x10: irq 11, address 00:10:a7:0b:16:ed rlphy0 at rl0 phy 0: RTL internal PHY Greetings, Hans
Re: mbuf leak with rl
El Jueves, 14 de Septiembre de 2006 17:38, escribiC3: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that mbufs > will slowly leak when using it. I noticed this after switching to 3.9. I > don't know if something happened to the card or not... maybe there is a > hardware error now that is making it behave funky. > > If you're using a "rl*" can you take a look at your mbuf usage (netstat > -m)? Me and another person both see something similar. > > Thanks, > Chris > > > dmesg: > rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 11, address > 00:48:54:65:39:5a > rlphy0 at rl0 phy 0: RTL internal PHY Look, I have a realtek NIC in OpenBSD 3.7 and OpenBSD 3.9: OpenBSD 3.9: # dmesg | grep rl rl0 at pci0 dev 13 function 0 "D-Link Systems 530TX+" rev 0x10: irq 11, address 00:0d:88:1a:8e:3a rlphy0 at rl0 phy 0: RTL internal PHY rlphy1 at vr0 phy 1: RTL8201L 10/100 PHY, rev. 1 # netstat -m 4 mbufs in use: 1 mbuf allocated to packet headers 3 mbufs allocated to socket names and addresses 0/10/6144 mbuf clusters in use (current/peak/max) 28 Kbytes allocated to network (3% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines # OpenBSD 3.7: # netstat -m 12 mbufs in use: 1 mbuf allocated to packet headers 11 mbufs allocated to socket names and addresses 0/64/6144 mbuf clusters in use (current/peak/max) 168 Kbytes allocated to network (1% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines # dmesg | grep rl rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 12 address 00:03:2d:04:60:40 rlphy0 at rl0 phy 0: RTL internal phy rl1 at pci0 dev 9 function 0 "Realtek 8139" rev 0x10: irq 10 address 00:03:2d:04:60:3f rlphy1 at rl1 phy 0: RTL internal phy rl2 at pci0 dev 10 function 0 "Realtek 8139" rev 0x10: irq 11 address 00:03:2d:04:60:3e rlphy2 at rl2 phy 0: RTL internal phy rl3 at pci0 dev 11 function 0 "Realtek 8139" rev 0x10: irq 15 address 00:03:2d:04:60:3d rlphy3 at rl3 phy 0: RTL internal phy What do you think? -- Abel TalaverC3n Estevez Ingeniero Superior de Telecomunicaciones Analista de Proyectos OpenWired Caballero 87 - Bajos 08029 - Barcelona Tel. 93 495 0990 Fax. 93 419 4591 http://www.openwired.com
Re: mbuf leak with rl
On Thursday 14 September 2006 17:38, Karle, Chris wrote: > Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that mbufs > will slowly leak when using it. I noticed this after switching to 3.9. I > don't know if something happened to the card or not... maybe there is a > hardware error now that is making it behave funky. > > If you're using a "rl*" can you take a look at your mbuf usage (netstat > -m)? Me and another person both see something similar. > > Thanks, > Chris > > > dmesg: > rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 11, address > 00:48:54:65:39:5a > rlphy0 at rl0 phy 0: RTL internal PHY OpenBSD 4.0 (GENERIC) #1104: Fri Sep 1 11:54:27 MDT 2006 [EMAIL PROTECTED]:/usr/src/sys/arch/i386/compile/GENERIC rl0 at pci0 dev 12 function 0 "Realtek 8139" rev 0x10: irq 11, address 00:0a:cd:06:2d:c1 rlphy0 at rl0 phy 0: RTL internal PHY rl1 at pci0 dev 13 function 0 "Realtek 8139" rev 0x10: irq 12, address 00:e0:7d:90:b2:22 rlphy1 at rl1 phy 0: RTL internal PHY 1513 mbufs in use: 1342 mbufs allocated to data 6 mbufs allocated to packet headers 165 mbufs allocated to socket names and addresses 363/590/6144 mbuf clusters in use (current/peak/max) 1616 Kbytes allocated to network (68% in use) 0 requests for memory denied 0 requests for memory delayed 0 calls to protocol drain routines 6:08PM up 11 days, 16 hrs, 4 users, load averages: 0.42, 0.28, 0.22 I DO see every once in a while isc-dhcp saying something along the lines "can't send - no buffer space available" - but that's on an interface served by ralink card. And no, I don't really know how to interpret the above data, I just hope it will be useful to someone ;) -- viq
mbuf leak with rl
Is anyone using a Realtek 8139 card with OpenBSD 3.9? I noticed that mbufs will slowly leak when using it. I noticed this after switching to 3.9. I don't know if something happened to the card or not... maybe there is a hardware error now that is making it behave funky. If you're using a "rl*" can you take a look at your mbuf usage (netstat -m)? Me and another person both see something similar. Thanks, Chris dmesg: rl0 at pci0 dev 8 function 0 "Realtek 8139" rev 0x10: irq 11, address 00:48:54:65:39:5a rlphy0 at rl0 phy 0: RTL internal PHY