Re: reliability of datagrams between processes on same machine?
- From: Maxwell Lol <nospam@xxxxxxxxxxx>
- Date: Sun, 04 Jan 2009 22:42:47 -0500
The Night Tripper <jkn_gg@xxxxxxxxxxxx> writes:
Hi David
Datagrams, by definition, don't provide this guarantee. In practice, a
system that did provide this guarantee would have to block in the case
where the send queue got full (or they'd have to provide an unlimited
amount of buffering).
OK, but how about reliable delivery if we assume that the send queue never
gets full? The processes are not doing huge amounts apart from
select()ing...
I was involved in a commerial real-time video project using Solaris,
and I did several tests to check reliability of UDP datagrams/multicasting.
For what it's worth - these things helped on Solaris 10 years ago.
1) Increase the socket buffer size.
2) Give the process low latency by using real time scheduling
3) Locking the process in memory so it didn't get swapped out.
.
- Follow-Ups:
- Re: reliability of datagrams between processes on same machine?
- From: The Night Tripper
- Re: reliability of datagrams between processes on same machine?
- References:
- reliability of datagrams between processes on same machine?
- From: The Night Tripper
- Re: reliability of datagrams between processes on same machine?
- From: David Schwartz
- Re: reliability of datagrams between processes on same machine?
- From: The Night Tripper
- reliability of datagrams between processes on same machine?
- Prev by Date: Re: skype and NAT
- Next by Date: Re: skype and NAT
- Previous by thread: Re: reliability of datagrams between processes on same machine?
- Next by thread: Re: reliability of datagrams between processes on same machine?
- Index(es):
Relevant Pages
|