nagle algorithm: any help here?

[Alan Riggs]

I got no responses in the networking forum, so I thought I would chance a cross-post. Any gurus here know whether the nagle algorithm is implemented by default for HP-UX 10.20 & 11.00?

[Curtis Larson]

in the defects section for the ARPA transport patch there is a reference to:

The sendfile system call can trigger the Nagle algorithm, decreasing file transfer performance.

being using the nagle algorith decreases performance and is considered a defect, my guess is that it doesn't use it by default.

[Alan Riggs]

Well, it isn't so clear to me. I read the same report, but it can also be interporeted to mean that the nagle algorithm i sin effect, and the arpa sendfile system can trigger (i.e. meet the small packet definition required for the nagle algorithm to delay packets) the effect. But I can see other readings (yours, for instance) as also possible.

I am hoping someone can confirm the truth one way or another.

[Berlene Herren]

By default telnetd calls setsockopt() TCP_NODELAY - immediately send data as it is received from the upper level. This means it does NOT use the Nagle Algorithm, and the tcp_naglim_def = 1. And at 11.0 this means
small one byte tcp packets are sent on the LAN.

You can prevent this by passing -TCP_DELAY to telnetd, which simply stops telnetd passing the setsockopt(TCP_NODELAY) flag to the transport.]

Berlene

[Alan Riggs]

Thanks, Berlene. Do you have the info, or can you point me to a doc, for other TCP protocols? In particular, I am interested in how sqlnet and remsh packets are handled.