hey guys,
as of now is there any software capable of tcp window manipulation for linux? or windows?

the only way i know of is to drop perfectly good packets, which isnt very efficient.

cheers

ET/BWMGR is what i have found? bandwidth manipulation software. but does it do tcp window manipulation? (dynamically??)

http://www.etinc.com/index.php?page=bwdemo.htm

anyone have any ideas?

Following are some of the more noteworthy features of the ET/BWMGR.

HTML Management Interface (GUI)
Bi-directional Bandwidth Limiting and Allocation
Multi-port Bandwidth Managment at full gigabit speeds
Unlimited Sessions or Policies (limited by CPU power only)
Integrated System Monitor (analyze network activity)
Integrated Firewall Function
Multi-Speed Bridging on Any Number of Ethernet Interfaces (ie bridge gigabit to multiple 100Mb/s)
Manage Traffic by IP, MAC or Network Address or Protocol
p2p Traffic Control (KaZaA, Morpheus, iMesh, AudioGalaxy, etc)
Virtual Name Host Limiting and monitoring
Reverse Limiting (dynamic policy creation based on a trigger policy)
Balanced Reverse Limiting (Each client gets proportionate chunk of bandwidth)
Hard Limits and Optional "Burst" Limit for Each Defined Policy
Flexible Bursting Policy Definitions
Hard or Automated TCP Rate Limiting (TCP Window manipulation)
"Weighted" Priorities and Guaranteed Minimum Bandwidth
Bandwidth Templates/Profiles Simplify Large-Scale Deployment
10 Priority Levels
Full VLAN (VPN) support
Advanced Group Functions (group ceilings, balanced groups, member overrides, nested groups)
Interface (port) Limits
Limit traffic in, out separately or limit total bandwidth used
Separate "rulesets" for Firewall, Bandwidth Management and Prioritization
Bypass Priority for Local Data Streams
Change Rules Dynamically From Simple Scripts
Integrated Source and Policy Routing (Use as a transparent proxy for caching, etc)
Statistics gathering and reporting for any definable traffic type (IP, MAC, VLAN, data type)
Statistics-only mode ("sniffs" your LAN and gathers stats without being inline)
Export Statistics via SNMP for any data type
Graphing of any definable address, network or traffic type with Daily, Weekly and Monthy Usage Reports
Graphing of external Routers and Devices with integrated SNMP Client.

As far as I've read, the only app what you ask for is Packeteer, and that's commercial. Maybe you could explain why you're focussing on that one issue as the main selection criterium?

On RH systems I no you can echo the TCP window high water level into some file in /proc. I can't remember I'll look it up and reply. This will allow you to increase it which should give you better performance when you're maxing out your connection. It also gives you worse performance for interactive processes. In the past I've barely noticed this but on a multi user system could be an issue.

Another thing I want to bring to your attention is the Web100 project which has some really cool real time instrumenting of TCP/IP as well as the ability to control the TCP/IP buffer in real time. The cavaet is that both clients and server need to have the Web100 patch. Could be an option. I know when I screwed around with this about a year and half ago I was able to push my transfers closer to the theoretical maxium than I'd ever been able to. It lets you visualize the latency and manually adjust for it.

-e

Hi,

I think the files that ealpert1 is referring to are:

/proc/sys/net/core/rmem_max
/proc/sys/net/core/rmem_default
/proc/sys/net/core/wmem_max
/proc/sys/net/core/wmem_max

I have redhat9 and they're already set to 64k

the reason i wanto to adjust this is because; no doubt with my 128k downstream adsl connection the buffer at the isp is full (causing huge latency) the only way i can fix this is by adjusting the tcp window size so that the senders of data slow it down so that the FIFO buffer at the ISP's router remains much lower.

this also helps with interactive connections such as telnet

I am having a look at the web100 site www.web100.org

there are also other projects like net100 built on web100 what is the difference?? anyone know i am reading about it, but if someone can put it in simple terms, that would be kool!

Also would web100 do a good job with my 128k adsl connection?

TCP/IP does this by itself. Part of the standard states that congestion control MUST be implemented to avoid this very thing.

The only reason you need to use these traffic engineering programs is if you want to artificially induce the congestion control algorith on certain types of traffic allowing other TCP sessions to have more bandwidth share.

More info here, if you're interested!

http://www.ietf.org/rfc/rfc2581.txt

Another option is to use tc. I do something like:

tc qdisc add dev eth0 handle ffff: ingress
tc filter add dev eth0 parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate 200kbit burst 10k drop flowid :1

This has the affect of dropping packets when they are arriving faster than 200 kbit

This may be useful http://www.tldp.org/HOWTO/Adv-Routing-HOWTO/index.html along with http://www.tldp.org/HOWTO/Bandwidth-...WTO/index.html if you want to make your cable modem alittle more efficient when it comes to interactive performance under load.

cheers

Jamie...