On Mar 28, 2011, at 15:27 Damon Permezel wrote:
> IIS accepts the "accept-encoding: gzip, deflate" and sends text/html data encoded. This means that the initial 12318 uncompressed entity body is in 4394 bytes for IIS and 12653 bytes for OSX, so the initial "index.html" comes back in less packets from IIS meaning less time, but that is not the whole story.
>
> For IIS, the entire conversation <snip>= 159,613
> For OSX, the entire conversation <snip>= 165,188
I see you are using a client tool (which one?).
Only the test files that the pingdom test loaded were distributed from the main web server to the test installation. And I also didn't bother to configure each webserver as identical as possible due to the large difference.
> Not a lot of difference, assuming I did not stuff something up.
>
> There is also an issue I take to mean that IIS/Windoze is defeating slow-start.
>
> From OSX, you see (arbitrary sec.tenths time aligned to the syn|ack for both):
>
> syn
> 2.9 syn|ack
> 2.9 ack
> 2.9 GET
> 3.4 ack (dup of ack in SYN|ACK above)
> 3.4 ack
> 3.4 RESP
> 3.4 ACK
> 3.9 RESP
> 4.0 ACK
> 4.4 RESP
> 4.4 RESP
> 4.5 ACK
>
> and on IIS:
>
> syn
> 2.9 syn|ack
> 2.9 ack
> 2.9 GET
> 3.3 RESP
> 3.3 RESP|ACK
> 3.4 ACK (dup of ACK above)
> 3.5 ACK
> 3.9 RESP
> 3.9 RESP
> 3.9 ACK
> 3.9 RESP
>
> IIS are on the front foot with the response. Looks to me to be slow-start. With 5 separate TCP sessions (both OSX and IIS) slow start adds up.
>
> For this particular TCP session (GET /images/snelheid-online-backup.png) it takes OSX 6.3 seconds total, and IIS 2.1 seconds total. Same # RESP packets, different timing due to what I claim appears to be traditional slow start.
This is interesting, and looks to me like the key to explaining these load time differences.
> Hmm. I remember now doing quick and dirty hacks to enable slow start defeat on TCP sessions, but it is past bed time, so I'll think no more on that.
Anyone else on how to defeat slow start TCP sessions on Mac OS X?
> Just a quick and dirty analysis when I should have gone off to bed 30 minutes ago, so it might all be crap.
For now I want to thank you.
> On 2011-Mar-26, at 10:22, Ceriel Jacobs wrote:
>> Google webmaster tools indicated slow page loading times for one of our static html websites (no php or other server side languages) running on a 2 x 2Ghz Dual-Core Intel Xeon Xserve with 7GB of RAM and less than 20% CPU load.
>>
>> <snip>
>> Even running IIS (and GWAN 1.0.5) on Windows 2003 in this VMware Fusion environment with only 368MB of memory and 1 processor assigned, the web page load time results in 0.34 seconds [2] and slower for full web page load time. Where the Host Mac OS X host OS is stuck at delivering in 1.03 seconds or slower. [1]
>> <snip>
>> Why does 10.5 Apache not serve static web content faster then 1.03s versus 0.34s for IIS/6.0 in a VM on the same machine?
>> What can be done to significantly shorten delivery times of static web content on Mac OS X build-in Apache web servers?
>> <snip>
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Macos-x-server mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden
