Re: Q: how to see/set the maximum network conections

From: John LeMay (jlemay@njmc.com)
Date: Tue Jan 25 2000 - 14:15:35 EST


I'm not sure what the Squid specific answer is, if there even is one,
however typical caching devices tend to drop connections or show a high
latency (for obvious reasons) once the throughput of the device (be it
software or asic) has been realized or exceeded. For additional general
information on caching performance, you may want to check out some of
the work I recently did for customers such as Compaq and HydraWeb during
my time with The Tolly Group.

One product I used and was especially impressed with was IRCache's Web
Polygraph (1.x). It provides an excellent simulation of clients
requesting docs from web servers and can report the data in a couple of
formats. Much better than using "stuff" like Netcom System's SmartTCP
for "session setup" measurements or ZDNet's WebBench which is a bit
inconsistent, IMHO.

>From the Tolly site (www.tolly.com), select "Tolly Verified" on the
right side of the page, register if you need to, and select all docs
sorted by date. All of my "web acceleration" work has been posted by
Tolly in the past week or so. You will notice that not all docs posted
recently relate to web caching, however.

Mathias Feiler wrote:
>
> -----BEGIN PGP SIGNED MESSAGE-----
>
> Hello all
>
> Context :
> We (Calculationcenter of University of Hohenheim Germany) set up
> a webcache using squit. We are testing it right now and works just fine
> for us.
> Now we think of open this new webcache to our customer. This mean we
> would increas the user from about 50 to about 5000 and more.
>
> hesitate and what we found:
> We hesitate if the maximum network connection is ok.
> So we checked the kerneldocs, but all we fond was a maximum number of
> ports useable (1024-4999).
> On the other side netstat showes that all the connection are made
> to the same port (of cause) AND REMAIN ON THAT PORT (3128).
> In other words we can have more open connection than used ports.
> netstat -a | grep bewbcache:3128 | wc -l
> showes that i.e. 81 connections are acting on port 3128
> Some in "TIME_WAIT" some in "ESTABLISHED"
>
> Question :
> If I can have more and more connections to/on the same port:
> What limit I'm running against preveting users open one more connection?
> How is it called?
> How can we tune this Parameter?
>
> By the way :
> What exactly means "TIME_WAIT"
>
> Thank you very much for answering our Questins.
>
> sincerily
>
> Mathias Feiler & Roland Hofmann
> - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
> Hochachtungsvoll und mit freundlichen Gruessen aus dem Dungeon :
> xxxxx MATHIAS
> X __ \ ____/ ___/ / / ___/ __
> X C O-O / / / / / / \ _\||/_
> X _\ __/ __/ / / __/ _ ___/ o o
> X _@ / / / / / / \ (_)
> | | __/ _____/ __/ _____/ _____/ __/ \__ ===
>
> M. Feiler Roßbergstr.1 72649 Wolfschlugen 0049 (0)7022 560965 (Privat)
> feiler@uni-hohenheim.de RZ (620) 0049 (0)711/459-3949 (Uni)
> PGP public key & Homepage : http://www.uni-hohenheim.de/~feiler
> - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
>
> -----BEGIN PGP SIGNATURE-----
> Version: 2.6.3i
> Charset: noconv
> Comment: Tell me where to get Your public key from!
>
> iQCVAwUBOI3fR3iYsAQ4godFAQGsLQP9F56pkA8T+NBRHVYHpEIno/OlV0x5lPz6
> c51PxzqFY8/P3abI3Wd9F3VWIdG7isOWXlUajItwK1bWj8Id1u6ZIqRpQPIKwIFh
> ngCrx8txi2aZUxO2daY23FQmpZLJfF/FcdQe3jUkemw+TJm/kEY6bvdsxvBWGncs
> fBCYbStuiDw=
> =V+PJ
> -----END PGP SIGNATURE-----
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-net" in
> the body of a message to majordomo@vger.rutgers.edu

-- 
John LeMay Jr.
Senior Enterprise Consultant
NJMC, LLC.

The rules have changed... Get paid to surf the web!!! http://www.alladvantage.com/go.asp?refid=bdc893


- To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.rutgers.edu



This archive was generated by hypermail 2b29 : Mon Jan 31 2000 - 21:00:32 EST