>
> Have you had any problems?...
>
There were some problems, especially with RAID and ENBD software.
First, I try to figure out how ENBD works, in fact when the NBD server
is going down what are the possibilities for the NBD client to know that
and "make an announcement" to RAID software and the last to unbind
the partition from its configuration. Of course, after a carefully read
of ENBD docs, I was able to deal with this problem.
One other problem (for me) was with linux kernel RAID support, because
you have to make a compromise between a small modularized kernel and
a "huge" speedy and reliable one. I choose the last.
>
>... What sort of throughput
> do you get? Have you had to actually do a rebuild? How much data are you storing?
>
The throughput depends of the NIC's you are using and it is
not dramatically limited to much by the software (RAID and ENBD).
There are some compares between NFS and ENBD in the ENBD docs
and you could see how fast is ENBD (it is fast, if you can trust me).
What I can say is:
- for reconstruction I am using 2 * 10BaseT NIC on each computer (3Com)
- both computers are running 2.2.16 kernel optimized
- one of those computers (the master) is PII-450MHz/64M RAM/IDE
and the other (the slave) is Pentium 100MHz/48M RAM/SCSI
- the partition for qmail is 1GByte large (small site)
In the consideration above, a full reconstruction it takes ~20min
or less, depending the load of the master which also running NS.
I think a rate of 5Mbit/sec it could be OK. It can be raised in multi-processor
configuration, more RAM on each nodes, SCSI on both, 100BaseT or 1GByte ethernet
and same architecture.
> Thanks,
> Rick.
It was my pleasure.
Regards,
--
Adrian Turcu
System Administrator
Computers Department
Romanian Railway Company
Constanta Region
E-mail: [EMAIL PROTECTED], [EMAIL PROTECTED]
Phone: +40 92 563791 (any time)
+40 43 363977 (home)
