Skip to contentSkip to conversation

How could I do this?

LARTC yangrunhua 3 participants 8 messages Dec 8, 2001 – Dec 9, 2001

lore
Share & tools
lore

yangrunhua

author
#0 lore raw

If I want to limit bandwidth from a lot of ip addresses( every ip has a limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to
the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy my demand that limit bandwidth for every ip, not for ip group), but it try to match line by line, then if many, the performance will go down.
Many thanks

#1 lore raw

IMHO u32 with hashes could be used for this. Other interesting way is that CBQ & HTB allows you to set packet's class from priority. When priority is 0x10003 then the packet is queued directly into 1:3 queue.
You can simly modify classifier (in fact I will do it for HTB) to allow such selection thru fwmark.
You can then mark flows by iptables .... --set-mark 0x10003 to assign packet into 1:3 class ..

HTH, devik

Quoted 12 lines · yangrunhua

If I want to limit bandwidth from a lot of ip addresses( every ip has a limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to
the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy my demand that limit bandwidth for every ip, not for ip group), but it try to match line by line, then if many, the performance will go down.
Many thanks

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

#2 lore raw
In reply to yangrunhua

Ehh sorry it is a lot of writting. Look at LARTC HOWTO 12.4 section. Basicaly if you have ip A.B.C.D then you can base hash source on D for several fixed A.B.C. Then you will end with at most N/256+1 lookups for N ip addresses.

devik

Quoted 41 lines · yangrunhua

Can you give me an example?
Thanks.

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Martin Devera
Sent: 2001Äê12ÔÂ8ÈÕ 15:58
To: yangrunhua
Cc: [email protected]
Subject: Re: [LARTC] How could I do this?

IMHO u32 with hashes could be used for this. Other interesting way is that CBQ & HTB allows you to set packet's class from priority. When priority is 0x10003 then the packet is queued directly into 1:3 queue.
You can simly modify classifier (in fact I will do it for HTB) to allow such selection thru fwmark.
You can then mark flows by iptables .... --set-mark 0x10003 to assign packet into 1:3 class ..

HTH, devik

Quoted 1 line · yangrunhua

If I want to limit bandwidth from a lot of ip addresses( every ip has

a

Quoted 6 lines

limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to
the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy

my

Quoted 2 lines

demand that limit bandwidth for every ip, not for ip group), but it try to match line by line, then if many, the performance will

go

Quoted 3 lines

down.
Many thanks

Quoted 6 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO:
http://ds9a.nl/2.4Routing/

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

#3 lore raw
In reply to yangrunhua
Quoted 4 lines · you

If I want to limit bandwidth from a lot of ip addresses( every ip has a limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to

Quoted 8 lines

the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy my demand t
hat limit bandwidth for every ip, not for ip group),
but it try to match line by line, then if many, the performance will go down.
Many thanks

I think the wrr qdisc can do this. It creates a class for each ip it sees. I never tried it, but I know some people who says it's working.

Stef

Signature · 8 lines
[email protected]
 More QOS info : http://docum.org/
 Title : "Using Linux as bandwidth manager"
     

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/
#4 lore raw
In reply to yangrunhua
Quoted 4 lines · you

If I want to limit bandwidth from a lot of ip addresses( every ip has a limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to

Quoted 8 lines

the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy my demand t
hat limit bandwidth for every ip, not for ip group),
but it try to match line by line, then if many, the performance will go down.
Many thanks

I think the wrr qdisc can do this. It creates a class for each ip it sees. I never tried it, but I know some people who says it's working.

Stef

Signature · 9 lines
[email protected]
 More QOS info : http://docum.org/
 Title : "Using Linux as bandwidth manager"
     


_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/
#5 lore raw
In reply to yangrunhua

I only wanted to mention it as interesting possibility but it can't be done in O(1). However it can be simply done in O(N^(1/M)) where M is integer > 1. You can do it by (untested):

ipchains -N sub
ipchains -A sub -s 0.0.0.1/0.0.0.255 -m +1 ipchains -A sub -s 0.0.0.2/0.0.0.255 -m +2 ..
..
ipchains -A input 192.168.1.0/24 -m 0x10000 -j sub ipchains -A input 192.168.2.0/24 -m 0x10100 -j sub ..
..

you would need 512 lines to handle 65536 adresses in 256 lookups on average. By creating another '-j subsub' level then you will end up with 96 rules for 65536 addresses with 60 lookups on average.

I didn't tested it, it is only idea. But should work.

hth, devik

Quoted 68 lines · yangrunhua

But how I design a filter( O(1) ) to classify based on fwmark(hash on fwmark)?

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of devik
Sent: 2001年12月8日 18:18
To: yangrunhua
Cc: [email protected]
Subject: RE: [LARTC] How could I do this?

Ehh sorry it is a lot of writting. Look at LARTC HOWTO 12.4 section. Basicaly if you have ip A.B.C.D then you can base hash source on D for several fixed A.B.C. Then you will end with at most N/256+1 lookups for N ip addresses.

devik

Quoted 41 lines · yangrunhua

Can you give me an example?
Thanks.

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Martin Devera
Sent: 2001Äę12ÔÂ8ČŐ 15:58
To: yangrunhua
Cc: [email protected]
Subject: Re: [LARTC] How could I do this?

IMHO u32 with hashes could be used for this. Other interesting way is that CBQ & HTB allows you to set packet's class from priority. When priority is 0x10003 then the packet is queued directly into 1:3 queue.
You can simly modify classifier (in fact I will do it for HTB) to allow such selection thru fwmark.
You can then mark flows by iptables .... --set-mark 0x10003 to assign packet into 1:3 class ..

HTH, devik

Quoted 1 line · yangrunhua

If I want to limit bandwidth from a lot of ip addresses( every ip has

a

Quoted 6 lines

limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to
the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy

my

Quoted 2 lines

demand that limit bandwidth for every ip, not for ip group), but it try to match line by line, then if many, the performance will

go

Quoted 3 lines

down.
Many thanks

Quoted 6 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO:
http://ds9a.nl/2.4Routing/

Quoted 5 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

#6 lore raw
In reply to yangrunhua

You can't mark packet with bandwidth and suppose that TC subsys will dynamically create shaped flows for you ;-) See other mail I send you. devik

Quoted 112 lines · yangrunhua

Thanks for your answer,
Now, I modified netfilter/iptables source,
I can mark every packet with some number attached with it's src ip(such as it's bandwidth); My question is: how can i limit every src ip's bandwidth with high effiency if I use tc?

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of devik
Sent: 2001年12月8日 23:12
To: yangrunhua
Cc: [email protected]
Subject: RE: [LARTC] How could I do this?

I only wanted to mention it as interesting possibility but it can't be done in O(1). However it can be simply done in O(N^(1/M)) where M is integer > 1. You can do it by (untested):

ipchains -N sub
ipchains -A sub -s 0.0.0.1/0.0.0.255 -m +1 ipchains -A sub -s 0.0.0.2/0.0.0.255 -m +2 ..
..
ipchains -A input 192.168.1.0/24 -m 0x10000 -j sub ipchains -A input 192.168.2.0/24 -m 0x10100 -j sub ..
..

you would need 512 lines to handle 65536 adresses in 256 lookups on average. By creating another '-j subsub' level then you will end up with 96 rules for 65536 addresses with 60 lookups on average.

I didn't tested it, it is only idea. But should work.

hth, devik

Quoted 68 lines · yangrunhua

But how I design a filter( O(1) ) to classify based on fwmark(hash on fwmark)?

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of devik
Sent: 2001嚴12月8旼 18:18
To: yangrunhua
Cc: [email protected]
Subject: RE: [LARTC] How could I do this?

Ehh sorry it is a lot of writting. Look at LARTC HOWTO 12.4 section. Basicaly if you have ip A.B.C.D then you can base hash source on D for several fixed A.B.C. Then you will end with at most N/256+1 lookups for N ip addresses.

devik

Quoted 41 lines · yangrunhua

Can you give me an example?
Thanks.

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Martin Devera
Sent: 2001Äę12ÔÂ8ČŐ 15:58
To: yangrunhua
Cc: [email protected]
Subject: Re: [LARTC] How could I do this?

IMHO u32 with hashes could be used for this. Other interesting way is that CBQ & HTB allows you to set packet's class from priority. When priority is 0x10003 then the packet is queued directly into 1:3 queue.
You can simly modify classifier (in fact I will do it for HTB) to allow such selection thru fwmark.
You can then mark flows by iptables .... --set-mark 0x10003 to assign packet into 1:3 class ..

HTH, devik

Quoted 1 line · yangrunhua

If I want to limit bandwidth from a lot of ip addresses( every ip has

a

Quoted 6 lines

limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to
the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy

my

Quoted 2 lines

demand that limit bandwidth for every ip, not for ip group), but it try to match line by line, then if many, the performance will

go

Quoted 3 lines

down.
Many thanks

Quoted 6 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO:
http://ds9a.nl/2.4Routing/

Quoted 5 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

Quoted 5 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/

#7 lore raw
In reply to yangrunhua

by the way I just found that if you attach fw filter without rules it will simply select class with classid == fwmark (nfmark in 2.4).
devik

Quoted 110 lines · yangrunhua

If I can modifiy skb->prio,
How can I use it to distribute packets directly to a class?

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of devik
Sent: 2001年12月8日 23:12
To: yangrunhua
Cc: [email protected]
Subject: RE: [LARTC] How could I do this?

I only wanted to mention it as interesting possibility but it can't be done in O(1). However it can be simply done in O(N^(1/M)) where M is integer > 1. You can do it by (untested):

ipchains -N sub
ipchains -A sub -s 0.0.0.1/0.0.0.255 -m +1 ipchains -A sub -s 0.0.0.2/0.0.0.255 -m +2 ..
..
ipchains -A input 192.168.1.0/24 -m 0x10000 -j sub ipchains -A input 192.168.2.0/24 -m 0x10100 -j sub ..
..

you would need 512 lines to handle 65536 adresses in 256 lookups on average. By creating another '-j subsub' level then you will end up with 96 rules for 65536 addresses with 60 lookups on average.

I didn't tested it, it is only idea. But should work.

hth, devik

Quoted 68 lines · yangrunhua

But how I design a filter( O(1) ) to classify based on fwmark(hash on fwmark)?

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of devik
Sent: 2001嚴12月8旼 18:18
To: yangrunhua
Cc: [email protected]
Subject: RE: [LARTC] How could I do this?

Ehh sorry it is a lot of writting. Look at LARTC HOWTO 12.4 section. Basicaly if you have ip A.B.C.D then you can base hash source on D for several fixed A.B.C. Then you will end with at most N/256+1 lookups for N ip addresses.

devik

Quoted 41 lines · yangrunhua

Can you give me an example?
Thanks.

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Martin Devera
Sent: 2001Äę12ÔÂ8ČŐ 15:58
To: yangrunhua
Cc: [email protected]
Subject: Re: [LARTC] How could I do this?

IMHO u32 with hashes could be used for this. Other interesting way is that CBQ & HTB allows you to set packet's class from priority. When priority is 0x10003 then the packet is queued directly into 1:3 queue.
You can simly modify classifier (in fact I will do it for HTB) to allow such selection thru fwmark.
You can then mark flows by iptables .... --set-mark 0x10003 to assign packet into 1:3 class ..

HTH, devik

Quoted 1 line · yangrunhua

If I want to limit bandwidth from a lot of ip addresses( every ip has

a

Quoted 6 lines

limit),
How could I improve performance( If I could use netfilter to mark the ip packet with the bandwidth assigned to
the src ip of packet), normally, this could only be done only by: one qdisc per ip, then there will be too many
filters to classify them based on fwmark(and u32 + hash can't satisfy

my

Quoted 2 lines

demand that limit bandwidth for every ip, not for ip group), but it try to match line by line, then if many, the performance will

go

Quoted 3 lines

down.
Many thanks

Quoted 6 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO:
http://ds9a.nl/2.4Routing/

Quoted 5 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

Quoted 5 lines

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/2.4Routing/

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/