Re: 10G with Intel card - GBIC options
* Andy a...@brandwatch.com [2013-12-02 10:05]: Henning, could you please confirm for us if the 32bit bandwidth limit was lifted in the new queuing subsystem, or if it is just still in place whilst dual-running the new and the old? otoh it is still there, but rather easy to lift. -- Henning Brauer, h...@bsws.de, henn...@openbsd.org BS Web Services GmbH, http://bsws.de, Full-Service ISP Secure Hosting, Mail and DNS Services. Dedicated Servers, Root to Fully Managed Henning Brauer Consulting, http://henningbrauer.com/
Re: 10G with Intel card - GBIC options
On Fri 24 Jan 2014 09:59:04 GMT, Henning Brauer wrote: * Andy a...@brandwatch.com [2013-12-02 10:05]: Henning, could you please confirm for us if the 32bit bandwidth limit was lifted in the new queuing subsystem, or if it is just still in place whilst dual-running the new and the old? otoh it is still there, but rather easy to lift. Cool ok. We'll wait until the next release when the old queuing system is removed allowing for a code clean up and this limit is lifted :) Thanks for your work, Andy.
Re: 10G with Intel card - GBIC options
On 2.12.2013. 10:05, Andy wrote: Hmm surprised by that! Henning, could you please confirm for us if the 32bit bandwidth limit was lifted in the new queuing subsystem, or if it is just still in place whilst dual-running the new and the old? I guess considering Hrvoje's findings the limit is still in place until ALTQ is removed completely in 5.5?? Cheers, Andy. Hi, second ix (82599) card is here and I have directly connected two servers. With kern.pool_debug=0, net.inet.ip.ifq.maxlen=1024 and mtu 16110 on ix cards bandwidth is ~7Gbps. tcpbench runs with -B 262144 -S 262144 pf.conf: set skip on lo block pass 10Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 10G max 10G queue bulk@ix0 parent queue@ix0 bandwidth 10G default pfctl -vsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 1G default qlimit 50 tcpbench shows 1404Mbps 9Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 9G max 9G queue bulk@ix0 parent queue@ix0 default bandwidth 9G pfctl -vsq queue queue@ix0 on ix0 bandwidth 410M, max 410M qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 410M default qlimit 50 tcpbench shows 206Mbps 8Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 8G max 8G queue bulk@ix0 parent queue@ix0 default bandwidth 8G pfctl -vsq queue queue@ix0 on ix0 bandwidth 3G, max 3G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 3G default qlimit 50 tcpbench shows 3690Mbps 7Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 7G max 7G queue bulk@ix0 parent queue@ix0 default bandwidth 7G pfctl -vsq queue queue@ix0 on ix0 bandwidth 2G, max 2G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 2G default qlimit 50 tcpbench shows 2695Mbps 6Gbps queue: queue queue@ix0 on ix0 bandwidth 6G max 6G queue bulk@ix0 parent queue@ix0 default bandwidth 6G pfctl -vsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 1G default qlimit 50 tcpbench shows 1699Mbps 5Gbps queue pf.conf queue queue@ix0 on ix0 bandwidth 5G max 5G queue bulk@ix0 parent queue@ix0 default bandwidth 5G pfctl -vsq queue queue@ix0 on ix0 bandwidth 705M, max 705M qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 705M default qlimit 50 tcpbench shows 218Mbps 4Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 4G max 4G queue bulk@ix0 parent queue@ix0 bandwidth 4G default pfctl -vsq queue queue@ix0 on ix0 bandwidth 4G, max 4G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 4G default qlimit 50 tcpbench shows 3986Mbps which is 99.65% of 4000Mbps. Could this 0,35% or 14Mbps bandwidth loss be interpret as queue overhead? If yes, then this is wonderful :) 3Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 3G max 3G queue bulk@ix0 parent queue@ix0 default bandwidth 3G pfctl -vsq queue queue@ix0 on ix0 bandwidth 3G, max 3G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 3G default qlimit 50 tcpbench shows 2988Mbps which is 0.40% bandwidth loss. 2Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 2G max 2G queue bulk@ix0 parent queue@ix0 default bandwidth 2G pfctl -vsq queue queue@ix0 on ix0 bandwidth 2G, max 2G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 2G default qlimit 50 tcpbench shows 1993Mbps which is 0,35% bandwidth loss 1Gbps queue: pf.conf queue queue@ix0 on ix0 bandwidth 1G max 1G queue bulk@ix0 parent queue@ix0 default bandwidth 1G pfctl -vsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 1G default qlimit 50 tcpbench show 996Mbps which is 0.7% bandwidth loss
Re: 10G with Intel card - GBIC options
On 2013-12-02, Kapetanakis Giannis bil...@edu.physics.uoc.gr wrote: I would love to go for the SFP+ path but we cannot afford it, flexoptix do a xenpak-sfp+ converter (and lots of other interesting things)
Re: 10G with Intel card - GBIC options
Hmm surprised by that! Henning, could you please confirm for us if the 32bit bandwidth limit was lifted in the new queuing subsystem, or if it is just still in place whilst dual-running the new and the old? I guess considering Hrvoje's findings the limit is still in place until ALTQ is removed completely in 5.5?? Cheers, Andy. On Fri 29 Nov 2013 22:10:20 GMT, Hrvoje Popovski wrote: On 29.11.2013. 17:08, Andy wrote: PS; I hope you have reeaaaly fast servers.. NB; ALTQ is currently 32bit so you cannot queue faster than 4 and a bit gig, unless you go for Hennings new queueing system which I'm still yet to do when I actually find time.. Hi, I'm not sure if new queueing system is faster than 4.3Gbps or pfctl -nvf pf.conf is lying or interface must be up and running to see real bandwith with pfctl -vvsq. I can't test it because I have one ix card. Will try to lend another ix card to see. # ifconfig ix0 ix0: flags=28843UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST,NOINET6 mtu 1500 lladdr 90:e2:ba:19:29:a8 priority: 0 media: Ethernet autoselect status: no carrier inet 10.22.22.1 netmask 0xff00 broadcast 10.22.22.255 pf.conf with 10G on ix0: queue queue@ix0 on ix0 bandwidth 10G max 10G queue ackn@ix0 parent queue@ix0 bandwidth 5G queue bulk@ix0 parent queue@ix0 bandwidth 5G default match on ix0 set ( queue (bulk@ix0i, ackn@ix0), prio (1,7) ) pfctl -nvf pf.conf queue queue@ix0 on ix0 bandwidth 1G, max 1G queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 705M queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 705M default pfctl -vvsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue ack@ix0 parent queue@ix0 on ix0 bandwidth 705M qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 705M default qlimit 50 pf.conf with 6G on ix0: queue queue@ix0 on ix0 bandwidth 6G max 6G queue ackn@ix0 parent queue@ix0 bandwidth 3G queue bulk@ix0 parent queue@ix0 bandwidth 3G default match on ix0 set ( queue (bulk@ix0i, ackn@ix0), prio (1,7) ) pfctl -nvf pf.conf queue queue@ix0 on ix0 bandwidth 1G, max 1G queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 3G queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 3G default pfctl -vvsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 3G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 3G default qlimit 50 pf.conf with 4G on ix0: queue queue@ix0 on ix0 bandwidth 4G max 4G queue ackn@ix0 parent queue@ix0 bandwidth 2G queue bulk@ix0 parent queue@ix0 bandwidth 2G default match on ix0 set ( queue (bulk@ix0i, ackn@ix0), prio (1,7) ) pfctl -nvf pf.conf queue queue@ix0 on ix0 bandwidth 4G, max 4G queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 2G queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 2G default pfctl -vvsq queue queue@ix0 on ix0 bandwidth 4G, max 4G qlimit 50 queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 2G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 2G default qlimit 50
Re: 10G with Intel card - GBIC options
On 29/11/13 19:16, Andy wrote: On Fri 29 Nov 2013 16:19:26 GMT, Kapetanakis Giannis wrote: Unfortunately on the Cisco part I don't SFP+. I have XENPACK option only which give me 3 options: SR ~ 3K GPL LRM ~ 1.5K GPL (I can't find any LRM GBIC for Intel side) CX4 ~ 600 GPL I'd avoid CX4, you wont find a CX4 NIC working well with OpenBSD nor would you want one tbh.. Stick with well known supported cards for OpenBSD.. Thanks for all the replies Andy. Are we totally sure about this? I'm talking about Intel - CX4 support on OpenBSD with ix(4). The manual page lists these: o Intel 82598EB 10GbE Adapter (10GbaseCX4) o Intel 82598EB Dual Port 10GbE Adapter (10GbaseCX4) o Intel 82599EB 10GbE Adapter (10GbaseCX4) Thanks Giannis
Re: 10G with Intel card - GBIC options
On Mon, Dec 02, 2013 at 11:36:31AM +0200, Kapetanakis Giannis wrote: On 29/11/13 19:16, Andy wrote: On Fri 29 Nov 2013 16:19:26 GMT, Kapetanakis Giannis wrote: Unfortunately on the Cisco part I don't SFP+. I have XENPACK option only which give me 3 options: SR ~ 3K GPL LRM ~ 1.5K GPL (I can't find any LRM GBIC for Intel side) CX4 ~ 600 GPL I'd avoid CX4, you wont find a CX4 NIC working well with OpenBSD nor would you want one tbh.. Stick with well known supported cards for OpenBSD.. Thanks for all the replies Andy. Are we totally sure about this? I'm talking about Intel - CX4 support on OpenBSD with ix(4). The manual page lists these: o Intel 82598EB 10GbE Adapter (10GbaseCX4) o Intel 82598EB Dual Port 10GbE Adapter (10GbaseCX4) o Intel 82599EB 10GbE Adapter (10GbaseCX4) CX4 should work fine but has mostly been replaced by SFP+ direct attach/copper and 10GBase-T with new cards.
Re: 10G with Intel card - GBIC options
Yea CX4 will work, its the chipset that matters. But CX4 is short range and superseded, and by using SFP+ you can pick and choose your transceivers for fibre or CAT cabling etc. On Mon 02 Dec 2013 10:10:37 GMT, Jonathan Gray wrote: On Mon, Dec 02, 2013 at 11:36:31AM +0200, Kapetanakis Giannis wrote: On 29/11/13 19:16, Andy wrote: On Fri 29 Nov 2013 16:19:26 GMT, Kapetanakis Giannis wrote: Unfortunately on the Cisco part I don't SFP+. I have XENPACK option only which give me 3 options: SR ~ 3K GPL LRM ~ 1.5K GPL (I can't find any LRM GBIC for Intel side) CX4 ~ 600 GPL I'd avoid CX4, you wont find a CX4 NIC working well with OpenBSD nor would you want one tbh.. Stick with well known supported cards for OpenBSD.. Thanks for all the replies Andy. Are we totally sure about this? I'm talking about Intel - CX4 support on OpenBSD with ix(4). The manual page lists these: o Intel 82598EB 10GbE Adapter (10GbaseCX4) o Intel 82598EB Dual Port 10GbE Adapter (10GbaseCX4) o Intel 82599EB 10GbE Adapter (10GbaseCX4) CX4 should work fine but has mostly been replaced by SFP+ direct attach/copper and 10GBase-T with new cards.
Re: 10G with Intel card - GBIC options
On 02/12/13 17:15, Andy wrote: Yea CX4 will work, its the chipset that matters. But CX4 is short range and superseded, and by using SFP+ you can pick and choose your transceivers for fibre or CAT cabling etc. Well the Cisco CX4 costs ~ 600$ List price, while the SR one costs 3.000$ List price. That's my main problem... I would love to go for the SFP+ path but we cannot afford it, so the CX4 seems like my only choice so far if it's ok with OBSD. G
Re: 10G with Intel card - GBIC options
The choice is of course yours.. ;) It would be worth trying a Cisco 'compatible' first before spending the big bucks on 'branded' optics.. http://www.gbics.com/xenpak-10gb-sr/?gclid=CKv_96G-irsCFSX4wgodQDEAdA Anyway, this is quite a personal decision and does affect support.. On Mon 02 Dec 2013 15:52:07 GMT, Kapetanakis Giannis wrote: On 02/12/13 17:15, Andy wrote: Yea CX4 will work, its the chipset that matters. But CX4 is short range and superseded, and by using SFP+ you can pick and choose your transceivers for fibre or CAT cabling etc. Well the Cisco CX4 costs ~ 600$ List price, while the SR one costs 3.000$ List price. That's my main problem... I would love to go for the SFP+ path but we cannot afford it, so the CX4 seems like my only choice so far if it's ok with OBSD. G
Re: 10G with Intel card - GBIC options
Kapetanakis Giannis [bil...@edu.physics.uoc.gr] wrote: On 02/12/13 17:15, Andy wrote: Yea CX4 will work, its the chipset that matters. But CX4 is short range and superseded, and by using SFP+ you can pick and choose your transceivers for fibre or CAT cabling etc. Well the Cisco CX4 costs ~ 600$ List price, while the SR one costs 3.000$ List price. That's my main problem... I would love to go for the SFP+ path but we cannot afford it, so the CX4 seems like my only choice so far if it's ok with OBSD. ebay for a cisco CX4 Xenpak for less than $100 USD
10G with Intel card - GBIC options
Hi, I've just received a Cisco 6704 for my 10G uplinks. I'm looking for a network adapter to put on my OpenBSD primary firewall. I had in mind to use Intel X520-SR2 but the SR module of Cisco is too expensive... So I'm looking for either LRM or CX4 GBIC options to put on the C6704. Has anyone connected to these transceivers with any quality Intel card? I can't find any LRM GBIC from intel. I found a CX4 one but the card is EOL. Thanks for any feedback. G
Re: 10G with Intel card - GBIC options
We bought the Intel x520-DA2 cards as they gives you the flexibility of using any SFP+ transceiver.. If you buy the SR2 you are locked to using short range fibre and the optics for the other end can get expensive! NB; Their is a whole world of compatible optics out there which are just as good but wont have support from the switch vendor and will need careful testing with your kit.. If you get the DA2 card, also buy a couple of the Intel optics which when added to the order cost the same as the SR2 and you end up with exactly the same product, but with the flexibility of being able to use 'SFP+ Direct Connect cables', SR optics, LR optics etc.. http://www.intel.com/support/network/adapter/pro100/sb/CS-030612.htm We bought the optics for the link to the WAN provider, and SFP+ Direct Connect cables for the link to our HP 8206zl switches. Intel cards support most of the official branded Direct Connect cables (don't get real cheap ones) so get Cisco branded as you have a Cisco switch.. Much cheaper than Cisco optics but the same end result. Andy. On Fri 29 Nov 2013 15:07:34 GMT, Kapetanakis Giannis wrote: Hi, I've just received a Cisco 6704 for my 10G uplinks. I'm looking for a network adapter to put on my OpenBSD primary firewall. I had in mind to use Intel X520-SR2 but the SR module of Cisco is too expensive... So I'm looking for either LRM or CX4 GBIC options to put on the C6704. Has anyone connected to these transceivers with any quality Intel card? I can't find any LRM GBIC from intel. I found a CX4 one but the card is EOL. Thanks for any feedback. G
Re: 10G with Intel card - GBIC options
PS; I hope you have reeaaaly fast servers.. NB; ALTQ is currently 32bit so you cannot queue faster than 4 and a bit gig, unless you go for Hennings new queueing system which I'm still yet to do when I actually find time.. On Fri 29 Nov 2013 16:05:35 GMT, Andy wrote: We bought the Intel x520-DA2 cards as they gives you the flexibility of using any SFP+ transceiver.. If you buy the SR2 you are locked to using short range fibre and the optics for the other end can get expensive! NB; Their is a whole world of compatible optics out there which are just as good but wont have support from the switch vendor and will need careful testing with your kit.. If you get the DA2 card, also buy a couple of the Intel optics which when added to the order cost the same as the SR2 and you end up with exactly the same product, but with the flexibility of being able to use 'SFP+ Direct Connect cables', SR optics, LR optics etc.. http://www.intel.com/support/network/adapter/pro100/sb/CS-030612.htm We bought the optics for the link to the WAN provider, and SFP+ Direct Connect cables for the link to our HP 8206zl switches. Intel cards support most of the official branded Direct Connect cables (don't get real cheap ones) so get Cisco branded as you have a Cisco switch.. Much cheaper than Cisco optics but the same end result. Andy. On Fri 29 Nov 2013 15:07:34 GMT, Kapetanakis Giannis wrote: Hi, I've just received a Cisco 6704 for my 10G uplinks. I'm looking for a network adapter to put on my OpenBSD primary firewall. I had in mind to use Intel X520-SR2 but the SR module of Cisco is too expensive... So I'm looking for either LRM or CX4 GBIC options to put on the C6704. Has anyone connected to these transceivers with any quality Intel card? I can't find any LRM GBIC from intel. I found a CX4 one but the card is EOL. Thanks for any feedback. G
Re: 10G with Intel card - GBIC options
On 29/11/13 18:05, Andy wrote: We bought the Intel x520-DA2 cards as they gives you the flexibility of using any SFP+ transceiver.. If you buy the SR2 you are locked to using short range fibre and the optics for the other end can get expensive! NB; Their is a whole world of compatible optics out there which are just as good but wont have support from the switch vendor and will need careful testing with your kit.. If you get the DA2 card, also buy a couple of the Intel optics which when added to the order cost the same as the SR2 and you end up with exactly the same product, but with the flexibility of being able to use 'SFP+ Direct Connect cables', SR optics, LR optics etc.. http://www.intel.com/support/network/adapter/pro100/sb/CS-030612.htm We bought the optics for the link to the WAN provider, and SFP+ Direct Connect cables for the link to our HP 8206zl switches. Intel cards support most of the official branded Direct Connect cables (don't get real cheap ones) so get Cisco branded as you have a Cisco switch.. Much cheaper than Cisco optics but the same end result. Andy. Unfortunately on the Cisco part I don't SFP+. I have XENPACK option only which give me 3 options: SR ~ 3K GPL LRM ~ 1.5K GPL (I can't find any LRM GBIC for Intel side) CX4 ~ 600 GPL I'm going probably for the CX4 option so I have to find an Intel server card that has CX4. My options for CX4 so far look like these: 82599EB which support all kinds of interfaces. I don't know if this is a server adapter or not (looks very cheap) Also it might need an CX4 interface to be attached. http://ark.intel.com/products/32207/Intel-82599EB-10-Gigabit-Ethernet-Controller Is 82599 chip the same on all the adapters? (EN, EB, X520, X540). Is it the same card with different optics? Also there is this NetEffect Ethernet Server Cluster Adapter CX4 http://ark.intel.com/products/55362/NetEffect-Ethernet-Server-Cluster-Adapter-CX4 but I don't know OpenBSD's support on this. Also it's PCIe v1.1 but I think this is my least problem. G ps. thanks for the reply
Re: 10G with Intel card - GBIC options
On Fri 29 Nov 2013 16:19:26 GMT, Kapetanakis Giannis wrote: On 29/11/13 18:05, Andy wrote: We bought the Intel x520-DA2 cards as they gives you the flexibility of using any SFP+ transceiver.. If you buy the SR2 you are locked to using short range fibre and the optics for the other end can get expensive! NB; Their is a whole world of compatible optics out there which are just as good but wont have support from the switch vendor and will need careful testing with your kit.. If you get the DA2 card, also buy a couple of the Intel optics which when added to the order cost the same as the SR2 and you end up with exactly the same product, but with the flexibility of being able to use 'SFP+ Direct Connect cables', SR optics, LR optics etc.. http://www.intel.com/support/network/adapter/pro100/sb/CS-030612.htm We bought the optics for the link to the WAN provider, and SFP+ Direct Connect cables for the link to our HP 8206zl switches. Intel cards support most of the official branded Direct Connect cables (don't get real cheap ones) so get Cisco branded as you have a Cisco switch.. Much cheaper than Cisco optics but the same end result. Andy. Unfortunately on the Cisco part I don't SFP+. I have XENPACK option only which give me 3 options: SR ~ 3K GPL LRM ~ 1.5K GPL (I can't find any LRM GBIC for Intel side) CX4 ~ 600 GPL I'd avoid CX4, you wont find a CX4 NIC working well with OpenBSD nor would you want one tbh.. Stick with well known supported cards for OpenBSD.. At a guess, I'd use an SR XENPACK on the Cisco side and connect to the Intel Optic Transceiver on the OBSD side, and use an LC to SC OM4 fibre etc.. NB; Try a 'Cisco compatible' XENPACK SR if you can't afford it, for example; http://www.gbics.com/xenpak-10gb-sr/?gclid=CKv_96G-irsCFSX4wgodQDEAdA http://www.intel.com/content/dam/doc/product-brief/ethernet-sfp-optics-brief.pdf Note the standards compatibility; 10GBASE-SR You will need to spend some money on a pair to try it (I don't know if these work but they look like they do), if it works get the others.. It'll either work perfectly or it wont at all.. Good luck, Andy I'm going probably for the CX4 option so I have to find an Intel server card that has CX4. My options for CX4 so far look like these: 82599EB which support all kinds of interfaces. I don't know if this is a server adapter or not (looks very cheap) Also it might need an CX4 interface to be attached. http://ark.intel.com/products/32207/Intel-82599EB-10-Gigabit-Ethernet-Controller Is 82599 chip the same on all the adapters? (EN, EB, X520, X540). Is it the same card with different optics? Also there is this NetEffect Ethernet Server Cluster Adapter CX4 http://ark.intel.com/products/55362/NetEffect-Ethernet-Server-Cluster-Adapter-CX4 but I don't know OpenBSD's support on this. Also it's PCIe v1.1 but I think this is my least problem. G ps. thanks for the reply
Re: 10G with Intel card - GBIC options
On 29.11.2013. 17:08, Andy wrote: PS; I hope you have reeaaaly fast servers.. NB; ALTQ is currently 32bit so you cannot queue faster than 4 and a bit gig, unless you go for Hennings new queueing system which I'm still yet to do when I actually find time.. Hi, I'm not sure if new queueing system is faster than 4.3Gbps or pfctl -nvf pf.conf is lying or interface must be up and running to see real bandwith with pfctl -vvsq. I can't test it because I have one ix card. Will try to lend another ix card to see. # ifconfig ix0 ix0: flags=28843UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST,NOINET6 mtu 1500 lladdr 90:e2:ba:19:29:a8 priority: 0 media: Ethernet autoselect status: no carrier inet 10.22.22.1 netmask 0xff00 broadcast 10.22.22.255 pf.conf with 10G on ix0: queue queue@ix0 on ix0 bandwidth 10G max 10G queue ackn@ix0 parent queue@ix0 bandwidth 5G queue bulk@ix0 parent queue@ix0 bandwidth 5G default match on ix0 set ( queue (bulk@ix0i, ackn@ix0), prio (1,7) ) pfctl -nvf pf.conf queue queue@ix0 on ix0 bandwidth 1G, max 1G queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 705M queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 705M default pfctl -vvsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue ack@ix0 parent queue@ix0 on ix0 bandwidth 705M qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 705M default qlimit 50 pf.conf with 6G on ix0: queue queue@ix0 on ix0 bandwidth 6G max 6G queue ackn@ix0 parent queue@ix0 bandwidth 3G queue bulk@ix0 parent queue@ix0 bandwidth 3G default match on ix0 set ( queue (bulk@ix0i, ackn@ix0), prio (1,7) ) pfctl -nvf pf.conf queue queue@ix0 on ix0 bandwidth 1G, max 1G queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 3G queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 3G default pfctl -vvsq queue queue@ix0 on ix0 bandwidth 1G, max 1G qlimit 50 queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 3G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 3G default qlimit 50 pf.conf with 4G on ix0: queue queue@ix0 on ix0 bandwidth 4G max 4G queue ackn@ix0 parent queue@ix0 bandwidth 2G queue bulk@ix0 parent queue@ix0 bandwidth 2G default match on ix0 set ( queue (bulk@ix0i, ackn@ix0), prio (1,7) ) pfctl -nvf pf.conf queue queue@ix0 on ix0 bandwidth 4G, max 4G queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 2G queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 2G default pfctl -vvsq queue queue@ix0 on ix0 bandwidth 4G, max 4G qlimit 50 queue ackn@ix0 parent queue@ix0 on ix0 bandwidth 2G qlimit 50 queue bulk@ix0 parent queue@ix0 on ix0 bandwidth 2G default qlimit 50
