[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]
Re: Aspera SCP software
- From: Brian Long <brilong cisco com>
- To: "Discussion of Red Hat Enterprise Linux 3 (Taroon)" <taroon-list redhat com>
- Subject: Re: Aspera SCP software
- Date: Mon, 16 May 2005 15:16:20 -0400
On Mon, 2005-05-16 at 17:36 +0100, John Haxby wrote:
> Ed Wilts wrote:
>
> >It does work well in practise. The wide-area SAN people use the same
> >concept. What appears to happen is that the transfers may go over UDP
> >instead of TCP so that you're not trying to ack every packet half-way
> >around the globe but can ack them in bunches. The end result appears to
> >be a more consistent data flow.
> >
> >
> This sounds like windowing to me or am I missing something? tcp(7)
> talks about setting the global read and write windows to deal with high
> bandwidth or high latency connections. With large windows you don't
> ack every packet, just every window'th packet don't you? (*)
In our internal transfers between two sites with ~60ms latency and
almost unlimited bandwidth (OC-12), you can only get about 1.4MB/sec on
single-threaded connections with default settings in RHEL 3. However,
if you add a multi-threaded rsync script in front of it, you can almost
saturate the WAN.
We use mdo, a tool written internally but recently open-sourced:
https://sourceforge.net/project/showfiles.php?
group_id=34678&package_id=140141
mdo is also named mscp, mrsh, etc and it's behavior changes depending on
how you call it. Of course, this is the opposite of what Ed wants to
do :)
/Brian/
--
Brian Long | | |
IT Data Center Systems | .|||. .|||.
Cisco Linux Developer | ..:|||||||:...:|||||||:..
Phone: (919) 392-7363 | C i s c o S y s t e m s
[Date Prev][Date Next] [Thread Prev][Thread Next]
[Thread Index]
[Date Index]
[Author Index]