Re: [lwip-users] slow SSI tag rendering

[address@hidden]

I saw that one, too, and fixed it 2 or 3 weeks ago. Although I don't remember if I checked it in to CVS already (I just came back from holidays so forgive me for forgetting that). The bug was that the httpd stopped sending after parsing SSI data and only sent again after a timeout. By fixing that and always sending until the window is full, TX speed went back to normal (like for other pages without SSI).

Please have a look at CVS HEAD.

Simon


Chris Gili wrote:

I am attempting to use the raw httpd with SSI support, but with lwip 1.3.2.  Using it on an ARM7TDMI with no OS.

 

I have a web page with 45 tags in it, and it renders very slowly, as in two-three tags per second.  Looking at wireshark, I don’t see any retransmits.  With lwip tcp debug messages, I do get a lot of “tcp_enqueue: queue too long 17 (16)” messages.  I tried increasing  TCP_SND_QUEUELEN as mentioned in some older lwip-users archives, but then I start running out of memory (“tcp_enqueue: could not allocate memory for zero-copy pbuf”).  So I question whether tweaking config parameters will solve my issues.

 

Are 45 tags simply too many for one page?  My ssi handler function is just a switch statement, and seems very quick even with printf’s at the start and the end:  both outputs appear virtually simultaneously, so I don’t think the lost time is there.  I could toggle a LED and put a scope on it, if someone wants actual timing info.

 

I’m reluctant to upgrade the entire lwip stack.  Additionally, I don’t know that this is even the reason for the slow rendering of the SSI page; other non-SSI pages render very quickly (no noticeable wait time).  Any suggestions if I’m heading down the correct path or not?  Any particular outputs that would help diagnose this issue?  Is this just not going to work without upgrading lwip, too?

 

Thanks,
Chris

 

_______________________________________________
lwip-users mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/lwip-users