From: Josef Bacik <jbacik@fb.com>
To: Andrei Borzenkov <arvidjaar@gmail.com>,
The development of GNU GRUB <grub-devel@gnu.org>,
<kernel-team@fb.com>
Subject: Re: [PATCH] tcp: add window scaling support
Date: Tue, 18 Aug 2015 10:48:20 -0700 [thread overview]
Message-ID: <55D36FE4.4010506@fb.com> (raw)
In-Reply-To: <55D20894.2010605@gmail.com>
On 08/17/2015 09:15 AM, Andrei Borzenkov wrote:
> 12.08.2015 18:57, Josef Bacik пишет:
>> Sometimes we have to provision boxes across regions, such as
>> California to
>> Sweden. The http server has a 10 minute timeout, so if we can't get
>> our 250mb
>> image transferred fast enough our provisioning fails, which is not
>> ideal. So
>> add tcp window scaling on open connections and set the window size to
>> 1mb. With
>> this change we're able to get higher sustained transfers between
>> regions and can
>> transfer our image in well below 10 minutes. Without this patch we'd
>> time out
>> every time halfway through the transfer. Thanks,
>>
>
> I do not think having 1M TCP window by default is good idea. We probably
> could make it configurable if there are evidences that it is really
> required.
Ok I'll add a config option.
>
>> Signed-off-by: Josef Bacik <jbacik@fb.com>
>> ---
>> grub-core/net/tcp.c | 42 +++++++++++++++++++++++++++++-------------
>> 1 file changed, 29 insertions(+), 13 deletions(-)
>>
>> diff --git a/grub-core/net/tcp.c b/grub-core/net/tcp.c
>> index 5da8b11..57a6d4e 100644
>> --- a/grub-core/net/tcp.c
>> +++ b/grub-core/net/tcp.c
>> @@ -106,6 +106,18 @@ struct tcphdr
>> grub_uint16_t urgent;
>> } GRUB_PACKED;
>>
>> +struct tcp_scale_opt {
>> + grub_uint8_t kind;
>> + grub_uint8_t length;
>> + grub_uint8_t scale;
>> +} GRUB_PACKED;
>> +
>> +struct tcp_synhdr {
>> + struct tcphdr tcphdr;
>> + struct tcp_scale_opt scale_opt;
>> + grub_uint8_t padding;
>> +};
>> +
>
> No, this does not scale. Just allocate large enough buffer and use
> pointers inside it. We may want to support multiple options at some point.
Fair enough.
>
>> struct tcp_pseudohdr
>> {
>> grub_uint32_t src;
>> @@ -555,7 +567,7 @@ grub_net_tcp_open (char *server,
>> grub_net_tcp_socket_t socket;
>> static grub_uint16_t in_port = 21550;
>> struct grub_net_buff *nb;
>> - struct tcphdr *tcph;
>> + struct tcp_synhdr *tcph;
>> int i;
>> grub_uint8_t *nbd;
>> grub_net_link_level_address_t ll_target_addr;
>> @@ -617,20 +629,24 @@ grub_net_tcp_open (char *server,
>> }
>>
>> tcph = (void *) nb->data;
>> + grub_memset(tcph, 0, sizeof (*tcph));
>> socket->my_start_seq = grub_get_time_ms ();
>> socket->my_cur_seq = socket->my_start_seq + 1;
>> - socket->my_window = 8192;
>> - tcph->seqnr = grub_cpu_to_be32 (socket->my_start_seq);
>> - tcph->ack = grub_cpu_to_be32_compile_time (0);
>> - tcph->flags = grub_cpu_to_be16_compile_time ((5 << 12) | TCP_SYN);
>> - tcph->window = grub_cpu_to_be16 (socket->my_window);
>> - tcph->urgent = 0;
>> - tcph->src = grub_cpu_to_be16 (socket->in_port);
>> - tcph->dst = grub_cpu_to_be16 (socket->out_port);
>> - tcph->checksum = 0;
>> - tcph->checksum = grub_net_ip_transport_checksum (nb, GRUB_NET_IP_TCP,
>> - &socket->inf->address,
>> - &socket->out_nla);
>> + socket->my_window = 32768;
>> + tcph->tcphdr.seqnr = grub_cpu_to_be32 (socket->my_start_seq);
>> + tcph->tcphdr.ack = grub_cpu_to_be32_compile_time (0);
>> + tcph->tcphdr.flags = grub_cpu_to_be16_compile_time ((6 << 12) |
>> TCP_SYN);
>> + tcph->tcphdr.window = grub_cpu_to_be16 (socket->my_window);
>> + tcph->tcphdr.urgent = 0;
>> + tcph->tcphdr.src = grub_cpu_to_be16 (socket->in_port);
>> + tcph->tcphdr.dst = grub_cpu_to_be16 (socket->out_port);
>> + tcph->tcphdr.checksum = 0;
>> + tcph->scale_opt.kind = 3;
>> + tcph->scale_opt.length = 3;
>> + tcph->scale_opt.scale = 5;
>> + tcph->tcphdr.checksum = grub_net_ip_transport_checksum (nb,
>> GRUB_NET_IP_TCP,
>> + &socket->inf->address,
>> + &socket->out_nla);
>>
>> tcp_socket_register (socket);
>>
>>
>
> There is one more place - grub_net_tcp_accept. Not sure if it ever be
> used though.
So I only care about connections we open, not about connections that are
opened to us which is why I did it this way. If we start to care about
how fast grub can transfer stuff _to_ somebody else we can add it there
as well. Thanks,
Josef
prev parent reply other threads:[~2015-08-18 18:21 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-08-12 15:57 [PATCH] tcp: add window scaling support Josef Bacik
2015-08-17 16:15 ` Andrei Borzenkov
2015-08-18 17:48 ` Josef Bacik [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=55D36FE4.4010506@fb.com \
--to=jbacik@fb.com \
--cc=arvidjaar@gmail.com \
--cc=grub-devel@gnu.org \
--cc=kernel-team@fb.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.