We are using a pac script, mostly with Windows XP clients ...

@Dean : the second page does not load faster than the first one. "Browser repsonse time" are much better early in the morning (with less connection obviously). When a blank page takes too much time to load, doing a "refresh" or revalidating the URL in the address bar often "unlock" page loading and the page is displayed with a good response time.

I'll try to investigate if DNS is involved and maybe find a workaround to the pac autoconfiguration to do transparent proxy.

Ionel


Ritter, Nicholas wrote:
 I had a problem similar to this at another job site a coulple of years ago. 
The clients were windows xp machines, and they were using wpad/pac style 
configuration. The fix was transparent caching.

-Nick


-----Original Message-----
From: Dean Weimer [mailto:[EMAIL PROTECTED] Sent: Thursday, June 05, 2008 1:53 PM
To: GARDAIS Ionel; [email protected]
Subject: [squid-users] Re:[squid-users] performances ... again

How are your browsers configured to use the proxy? Manual, wpad script, 
transparent?
Could be a problem with discovering proxy settings.

What about a second page on the same server, ie. http://some.domain.com then 
http://some.domain.com/nextpage.html?  Could be a DNS response issue, perhaps 
your first server is timing out, and the clients have to wait for the second to 
respond.  If the second page comes up right away, this would be a good 
indicator of that.  As Squid would have cached the DNS lookup from the first 
request.

Most servers are not going to have a 12Mb/s of bandwidth is a decent chunk, I 
wouldn't expect to see that maxed out all the time, because you are averaging 
under 2Mb/s in itself is not cause for concern.  The fact that you are hitting 
it on large downloads means the link is performing well.

I am seeing about 180ms median response time on misses and 5ms median response 
time on hits, 87ms response time on DNS Lookups.  The server is running 2G cpu 
and 1G ram, with an average of 900 req/min.  The server is servicing about 500 
clients connected behind 2 T1 lines.  Both lines are consistently running at 
1.2 to 1.5Mb/s from 7am to 6pm when most users are at work.  Disk cache is 
8gigs on the same disk as system, which is actually a hardware mirrored ultra 
160 10K SCSI disks, (Not ideal, as I have learned a lot more since I first 
built this system), but the performance is excellent, so I haven't found cause 
to change it.  The server is running FreeBSD 5.4, squid the cache and logs are 
installed on their own mount point using ufs file system, Mount point is on a 
single Disk slice encompassing entire hard drive, and to top that off, the file 
system runs about 90% of capacity, yet another no no.

Thanks,
     Dean Weimer
     Network Administrator
     Orscheln Management Co

-----Original Message-----
From: GARDAIS Ionel [mailto:[EMAIL PROTECTED]
Sent: Thursday, June 05, 2008 12:11 PM
To: chris brain; [email protected]
Subject: [squid-users] RE : [squid-users] Re:[squid-users] performances ... 
again

Hi Chris,

The internet link is not congested.
As I wrote, we use less than 2Mb/s of the 12Mb/s we can reach (but yes, upload 
seems to be limited to 512Kb/s (somewhere around the maximum of the line), this 
might be a bottleneck).
When downloading large files (from ten to hundereds of megabytes), the whole 
12Mb are used (showing a 1100KB/s download speed).

After rereading my post, I saw that I did not finish a line :
"[...] cache-misses median service times are around 200ms and cache-hits are around 
3ms" but we often see a 10-second lag for browser to start loading the page.

Ionel


-----Message d'origine-----
De : chris brain [mailto:[EMAIL PROTECTED]
Envoyé : jeudi 5 juin 2008 18:34
À : [email protected]
Objet : [squid-users] Re:[squid-users] performances ... again

Hi Ionel,

Your performance dont look that bad. Our stats roughly work out to be :

1000+ users
NTLM auth
Average HTTP requests per minute since start:   2990.8
with max 30% hits. (so your hits look coparable to us.) Our cache miss service 
time averages to about 160ms and cache hits service time about 10ms running IBM 
blade P4 3G cpu 1Gb ram. mirrored drive.

Our links can get quite congested and we dont get complaints about the 
performance.

Are you having internet link performance issues?? are you monitoring it
(snmp/netflow) ?

chris


------------------------------------------------------------------------------------
West Australian Newspapers Group
------------------------------------------------------------------------------------
Privacy and Confidentiality Notice

The information contained herein and any attachments are intended solely for 
the named recipients. It may contain privileged confidential information.  If 
you are not an intended recipient, please delete the message and any 
attachments then notify the sender. Any use or disclosure of the contents of 
either is unauthorised and may be unlawful. Any liability for viruses is 
excluded to the fullest extent permitted by law.

Advertising Terms & Conditions
Please refer to the current rate card for advertising terms and conditions.  
The rate card is available on request or via www.thewest.com.au

Unsubscribe
If you do not wish to receive emails such as this in future please reply to it with 
"unsubscribe" in the subject line.

--
Ionel GARDAIS
System-Network Engineer

begin:vcard
fn:Ionel GARDAIS
n:GARDAIS;Ionel
org:Tech'Advantage;IT
adr:;;1 Rue Isabey;Rueil Malmaison;FR;92500;FR
email;internet:[EMAIL PROTECTED]
tel;work:+33(0)147088131
tel;fax:+33(0)147088065
x-mozilla-html:FALSE
url:http://www.tech-advantage.com
version:2.1
end:vcard

Reply via email to