RE: [lwip-users] Sizes

[John Kennedy]

Bill,
That’s good to know.  BTW which checksum algorithm did base your assembly code 
on?  Does anyone know which of the checksum algorithms is the fastest? I'm a 
little surprised that other than the call overhead that an assembly version of 
memcpy is much faster than what the complier produces from the library?

John

-----Original Message-----
From: Bill Auerbach [mailto:address@hidden
Sent: Wednesday, May 27, 2009 2:45 PM
To: 'Mailing list for lwIP users'
Subject: RE: [lwip-users] Sizes

The biggest bang for my buck was to use assembly code for the inet_checksum and 
a hybrid inline/assembly code memcpy for all memcpy's used.  If your hardware 
does DMA memory-to-memory copies definitely use it for memcpy.

Bill

>-----Original Message-----
>From: address@hidden
>[mailto:address@hidden On
>Behalf Of John Kennedy
>Sent: Wednesday, May 27, 2009 4:12 PM
>To: Mailing list for lwIP users
>Subject: RE: [lwip-users] Sizes
>
>Jeff,
>Thanks for the info.  As it turns out there is more info about this
>issue in the archives, but it takes a little searching to dig it out.  I
>searched on TCP_MSS and got quite a bit of stuff... still looking.  Like
>you I came up with an TCP_MSS of 1460, a MEM_SIZE of 0x10000,
>TCP_SND_BUFF of 8 * TCP_MSS. But I set my TCP_SEND_QUEUELEN to 32 and my
>TCP_WND to 16384.  I learned from the archive that TCP_WND affects the
>TCP receive throughput.
>
>The archives also seem to indicate that a certain amount of port
>specific tuning for TCP is necessary.  So it would be helpful if someone
>could suggest some standard LWIP test scenarios along with some
>guidelines of what to tweak to determine the best configuration for
>one's particular needs.  Otherwise it's a lot of trial and error.
>
>So far Lwip has worked well for me (no bugs or problems except those of
>my own making) but I'm still fooling around with the configuration to
>maximize performance, and like you my port is very young.
>
>John
>
>
>
>________________________________________
>John Kennedy
>
>
>Idaho Technology Inc.
>390 Wakara Way
>Salt Lake City, UT 84108, USA
>
>USA: 1-800-735-6544
>Bus:+1 (801)736-6354 x448
>Fax:+1 (801)588-0507
>
>http://www.idahotech.com/
>-----Original Message-----
>From: Jeff Barber [mailto:address@hidden
>Sent: Wednesday, May 27, 2009 1:39 PM
>To: Mailing list for lwIP users
>Subject: Re: [lwip-users] Sizes
>
>On Wed, May 27, 2009 at 9:47 AM, John Kennedy
><address@hidden> wrote:
>> I'm using Lwip with sockets on MicroBlaze.  I have an MTU of 1500.  I
>want
>> to maximize the Ethernet TX/RX data rates (bits/sec) in my embedded
>system.
>> Memory is not an issue (512Mbytes avail).
>>
>> How does one determine the values for:
>>
>> *         MEM_SIZE
>>
>> *         PBUF_POOL_SIZE
>>
>> *         PBUF_POOL_BUFSIZE
>>
>> *         TCP_MSS
>>
>> *         TCP_SND_BUF
>>
>> *         TCP_SND_QUEUELEN
>>
>>  to maximize the Ethernet TX/RX data rates?
>
>I wish there were a better source of information on this as well.  I
>spent quite a while trolling through the source trying to understand
>what all the different settings are for.  Some I figured out, others
>I'm still not sure of.
>
>Like you, my environment is not memory-constrained.  Furthermore, I am
>using my own malloc implementation rather than the lwip built-in
>mem_malloc and memp_malloc (this is partially because I trust my own
>allocation functions and partially because I had a hard time
>understanding how to configure the lwIP implementation :-).  As best I
>can tell, when using one's own allocation routines, several of the
>memory-related defines are "don't care".  So MEM_SIZE and all the
>MEMP_* and PBUF_POOL_SIZE are (I think) not actually used -- beyond
>the sanity tests in lwip/src/core/init.c.
>
>Caveat: my port is very young -- my driver came up for the first time
>a couple of days ago, and it hasn't yet done much more than answer
>ping packets and ARP (no TCP yet).  FWIW, though, here's my lwipopts.h
>(for everything else, I accept the default values from <lwip/opt.h>):
>
>#define NO_SYS                      1
>#define MEM_LIBC_MALLOC             1
>#define MEMP_MEM_MALLOC             1
>#define MEM_ALIGNMENT               4
>#define MEM_SIZE                    (4 * 1024 * 1024)
>#define MEMP_NUM_PBUF               1024
>#define MEMP_NUM_UDP_PCB            20
>#define MEMP_NUM_TCP_PCB            20
>#define MEMP_NUM_TCP_PCB_LISTEN     16
>#define MEMP_NUM_TCP_SEG            128
>#define MEMP_NUM_REASSDATA          32
>#define MEMP_NUM_ARP_QUEUE          10
>#define PBUF_POOL_SIZE              512
>#define LWIP_ARP                    1
>#define IP_REASS_MAX_PBUFS          64
>#define IP_FRAG_USES_STATIC_BUF     0
>#define IP_DEFAULT_TTL              255
>#define IP_SOF_BROADCAST            1
>#define IP_SOF_BROADCAST_RECV       1
>#define LWIP_ICMP                   1
>#define LWIP_BROADCAST_PING         1
>#define LWIP_MULTICAST_PING         1
>#define LWIP_RAW                    0
>#define TCP_WND                     (4 * TCP_MSS)
>#define TCP_MSS                     1460
>#define TCP_SND_BUF                 (8 * TCP_MSS)
>#define TCP_LISTEN_BACKLOG          1
>#define LWIP_NETIF_STATUS_CALLBACK  1
>#define LWIP_NETIF_LINK_CALLBACK    1
>#define LWIP_NETIF_HWADDRHINT       1
>#define LWIP_NETCONN                0
>#define LWIP_SOCKET                 0
>#define LWIP_STATS_DISPLAY          1
>#define ETHARP_TRUST_IP_MAC         0
>
>#define mem_init()
>#define mem_free                    my_free
>#define mem_malloc                  my_malloc
>#define mem_calloc(c, n)            my_zalloc((c) * (n))
>#define mem_realloc                 my_realloc
>
>
>For the MSS value, specifically, I worked backwards from the standard
>max ethernet frame size: 1514.  Subtract 14 bytes for layer 2 header;
>subtract 20 bytes for IP header and another 20 for TCP header; that
>leaves 1460.  There are a couple of ports in contrib/ that define it
>to 1476 but I couldn't figure out how they came up with that.  I will
>have a direct ethernet connection to my peer so don't need to worry
>about any additional fragmentation.
>
>BTW, any comments on my settings would be greatly appreciated.
>
>Hope this helps.
>Jeff
>
>
>_______________________________________________
>lwip-users mailing list
>address@hidden
>http://lists.nongnu.org/mailman/listinfo/lwip-users
>
>________________________________________
>CONFIDENTIALITY NOTICE: This E-mail and any attachments are confidential
>information of the sender and are for the exclusive use of the intended
>recipient. If you are not the intended recipient, be aware that any
>disclosure, copying, distribution, or use of this E-mail or any
>attachment is prohibited. If you have received this E-mail in error,
>please notify us immediately by returning it to the sender and delete
>this copy from your system. Thank you for your cooperation.
>
>
>
>
>_______________________________________________
>lwip-users mailing list
>address@hidden
>http://lists.nongnu.org/mailman/listinfo/lwip-users



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

________________________________________
CONFIDENTIALITY NOTICE: This E-mail and any attachments are confidential 
information of the sender and are for the exclusive use of the intended 
recipient. If you are not the intended recipient, be aware that any disclosure, 
copying, distribution, or use of this E-mail or any attachment is prohibited. 
If you have received this E-mail in error, please notify us immediately by 
returning it to the sender and delete this copy from your system. Thank you for 
your cooperation.