[distcc] LZO compression

Jean-Eric Cuendet jean-eric.cuendet at linkvest.com
Tue Dec 17 11:41:00 GMT 2002


PS: BTW, I think that you should change mailman settings so Reply reply 
to the list, not to the sender only.

>So the question is whether the marginal cost of doing our own digest
>is so great as to make it worth varying the protocol and code path
>depending on whether ssh is used or not.  I'd prefer simplicity unless
>there's a measurable difference.
>
My thoughts is that distcc should be kept simple.
SSH overhead is *enormous* compared to simple checksumming!
It should be 1% overhead, not much. Opening an SSH channel means:
- Validating server with public Crypto
- Validating client with public Crypto
- Authenticating client user with public Crypto
- Exchanging private keys
- Crypto of all data
That is *a lot* of work!

I think that checksum should always be done, while compression should be 
optional.
When running with SSH transport, we could be able to use SSH own 
compression or not (by a DISTCC_SSH_OPTIONS for example) or distcc own 
compression (which should be faster since SSH one is gzip based)

>There might be some value in using an integrity check even for
>tunnelled connections.  For example, I know some people use rsh (which
>doesn't have a md) for speed inside clusters, and they might want to
>use distcc.  There have been ssh bugs that caused streamed connections
>to get corrupted, though I think not recently.
>
Yes, for RSH, Checksum+Compression options would be cool and useful.
-jec






More information about the distcc mailing list