As discussed earlier, in order to address the disk space shortage on adaql2, we wanted to attach the adaql1 disk system with the four adaql1/data disks to adaql2.
Unfortunately, it turned out that that is not as easy as thought: adaql2 has only one external SCSI bus. This bus can handle a maximum of 15 devices. With the extra disk system, there would be 16 devices total (8 RAID-1 arrays (mirrors)). We could try to take out one of the arrays, but that would mean a lot of reconfiguration. Also, we have not tested 15 disks on a single bus. There may be stability issues (although unlikely, given that the disk chassis have signal repeaters). Even though we have SCSI add-on cards, the one available add-on slot is already taken by the graphics card. Finally, it turns out that the software RAID arrays may not take well to being moved between hosts, especially since the md device numbers will clash. In the interest of stability, we should not proceed with this in the middle of the experiment.
One possible solution to both speed and space problems could be to alternate running the DAQ between adaql1 and adaql2. Once the disks on one host fill up, the shift crew would switch to the other machine. This assumes that adaql1 really is fully configured and ready for DAQ.
For the moment, I am addding a quick band-aid: I will mount the adaql1 disks on adaql2 via the 192 subnet. This should eliminate any network contention on the (presumably very busy) outgoing direction of the 164 interface on adaql2. This change is completely transparent.