The replication job ran last night in 2 hours. The problem was a router that I had been told to used for my servers. I noticed using traceroute that my desktop didn’t use ut, nor did the Linux boxes I had been playing with, and they all had very fast transfer speeds. So I tested yesterday afternoon with another Solaris box, and the speed jump was astounding. Now the network folks are trying to figure out why that router was such a bottleneck.
Think good thoughts, I may have figured this puppy out. More tomorrow.
The original problem is here. This morning, we moved the Sun box out of the DMZ and tested the speed there.
FTP inside the DMZ: 120 KBytes/s
FTP outside the DMZ: 7045 KBytes/s
Also, scp runs at about the same speed as ftp.
There’s something screwy about the DMZ router, I think.
I’m having a strange problem with FTP. I have a Solaris 9 box in a DMZ. FTP from my Solaris 9 boxes inside the firewall to the Solaris 9 box in the DMZ is going extremely slow – 60-100 Kbps. FTP from Windows boxes or Linux boxes inside the firewall to the Solaris 9 box in the DMZ is much, much faster – 4000 Kbps. FTP from Windows or Lunix boxes to a Windows server in the DMZ is also much faster, in the neighborhood of 4000 Kbps. It’s only the Solaris-to-Solaris transfer that’s slow. The network folks did some reconfiguring of the network last week, and maybe there’s some config parameter that Solaris needs to know about, but what? Can anybody give me a rational explanation for this?
Here’s an example – I have one roughly 3 GB file that I need to move between the two Solaris boxes. It takes about an hour to make the transfer. However, if I ftp it from the Solaris box behind the firewall to my PC (also behind the firewall), then from my PC to the Solaris box in the DMZ, the total for both transfers is about 10 minutes. Or I can transfer that file from one Solaris box to another one not in the DMZ in the same 10 minutes. It’s just between one Solaris box behind the firewall to the Solaris box inside the DMZ that’s exceedingly slow.
In my early days of Unix, 1985 or thereabouts, when I needed to create an archive I used cpio. I don’t really remember why I used cpio instead of tar. Tar seemed to be the ubiquitous Unix archiving command. I ran across few fellow cpio-ers. So at some point, I gave in and started using tar. I rarely use cpio now, and almost never run across a reference to it. So I find it curious that Oracle, in the midst of distributing their patchsets for Oracle 9i (9.2), suddenly switched from distributing as tarballs, to distributing as cpio archives. It’s like a flashback to the Reagan Administration.