Ahhh but you can have more than one in your dsm.sys file (under unix,
under other platforms it will be a command line option)
Here is a unix example...
dsm.sys file
SErvername dsmserv01
COMMmethod tcpip
TCPPort 1500
TCPServeraddress dsmserv01.amoco.com
(or TCPServeraddress 149.180.10.20 )
SErvername dsmserva1
COMMmethod tcpip
TCPPort 1500
TCPServeraddress dsmserva1.amoco.com
(or TCPServeraddress 149.180.20.20)
say you have 200 oracle data base files you have to back up as fast as
possible. The files are named ORADB000 thru ORADB199 in the directory
/oracle and are each 1 GB in size for a total of 200 GB.
You could say "dsmc archive '/oracle/ORADB*' -pass=pswd" and this
single process would move about 3-4GB/hr this would tape around 57
hours...
Knowing that a fddi will be max'ed out by about 9 concurrent adsm
sessions that are sending single, large files (500 MB to 2 GB) you
could run a script to do something like: (*note the serv name change*)
nohup dsmc archive '/oracle/ORADB00*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB01*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB02*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB03*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB04*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB05*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB06*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB07*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB08*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB09*' -serv=dsmserv01 -pass=pswd &
nohup dsmc archive '/oracle/ORADB10*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB11*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB12*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB13*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB14*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB15*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB16*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB17*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB18*' -serv=dsmserva1 -pass=pswd &
nohup dsmc archive '/oracle/ORADB19*' -serv=dsmserva1 -pass=pswd &
I would expect to see about 30-ish GB per hour down each fddi path for
a total traffic rate of 50+ish GB/hr thus turning our 57 hour archive
into a "less than 4 hour" archive of our data base!
Yes, I'll be happy with only 1% of your raise after you work this
miracle at your company... ;-)
naturally I can't give out our script that puts table spaces into
backup mode, initiates archives, watches for the number of active
dsmc's and keeps that number high while there is still work to be
done, etc... Hey, the DBA's need to have someting to do!
Oh, but don't run a DB backup, expiration, migration, or reclamation
during this process... they'll take your raise away...
later
Dwight
______________________________ Reply Separator _________________________________
Subject: Re: adsm and dual NICs
Author: CCannam (CCannam AT GENAM DOT COM) at unix,mime
Date: 6/4/98 8:53 AM
Dwight:
Because the network folks gave ADSM a specific IP address from the very
beginning, this is what is "coded" in the dsm.opt file on each client. I'd
be curious to see how you managed to make the client "look" for a specific
NIC in your options file, if in fact, that's where the information starts.
Sounds more like NIC configuration at install would do the trick, making
ADSM find the NIC that pointed to the destination server address --- that
would eliminate problems with internal tables, etc.
C.L.Cannam
Storage Management
GENAM/St. Louis, MO/USA
ccannam AT genam DOT com
Dwight Cook <decook AT AMOCO DOT COM> on 06/03/98 04:31:53 PM
Please respond to "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
cc: (bcc: Cindy Cannam)
Subject: Re: adsm and dual NICs
Ahhh sorry I didn't speak up a year ago... kind'a busy back then....
doing things along those lines....
OK, currently we'll run 100 Mb fddi's (multiple) Some client
machines will have 1 for user interface and 2 for adsm backup use...
Our adsm servers will have 2 or 3 fddi's. Naturally backups will go
faster if they stay on the same subnet. The interfaces each have
different IP addresses... the names will be something like DSMSRV01,
DSMSRVA1, DSMSRVB1 indicating different interfaces into the same adsm
server... then the next server might be DSMSRV02, DSMSRVA2...
Now this is the hard part... getting the individual client node
admin's to perform nslookup on the available names, determine which
name is the closest via which interface out of their box, making the
adsm parameters match that, then get their (the client machine)
routing info correct to use the desired path.
Yes, once there was actually a client with a network card on a same
subnet as an adsm server BUT they specified the incorrect server name
and caused the backups to go out the wrong interface and take a 3
router hop over to the adsm server. I discovered this through random
samples of IP addresses of things comming in with a local port 1500.
(and they were wondering why their backups were taking so long!)
YES! and you can even run multiple, concurrent backup/archive tasks
and push them over different networks to minimize the time required
for the activities... We've pushed as many as 18 concurrent archives,
9 down each of two network interfaces.
And YES, FWDIFF scsi I/O on the client can become the bottle neck.
That was one of the GREATER MOMENTS in time... when they still asked
"how can we make it go faster?" the answer was YOU HAVE TO BUY MORE
DISKS, MORE DISK CONTROLLERS AND PUT FEWER DISKS BEHIND EACH !
later,
Dwight
______________________________ Reply Separator
_________________________________
Subject: adsm and dual NICs
Author: CCannam (CCannam AT GENAM DOT COM) at unix,mime
Date: 6/3/98 2:38 PM
To all:
I know I've asked this question before, but it was some time ago (last year
sometime), and the response was rather small (because nobody was really
doing what I was asking about). But I believe with the recent introduction
of some of the more intelligent NICs out there (General Signal's LINK/9000
and FileSpeed from CNT for instance), there may be more people who are
employing this technology with ADSM.
The network folks have decided (wow! a decision was actually made) that
ADSM should have its own dedicated backbone (probably 100 Mb switched
Ethernet). As such, the need for dual NICs in each of the clients becomes a
reality. Assuming that each client has available slots for a 2nd NIC
(always a question with older clients), and that the network can be built
in a reasonable amount of time (out of my control), am I correct in telling
these people (1) that ADSM has no control over what NIC is used (I've never
seen anything in the documentation or on any screen for control designation
for a network card); (2) that the controls used will have to be configured
at the time that the 2nd NIC is installed and discovered by the existing
client; and (3) that as long as the NIC is correctly configured for ADSM
traffic only (assuming again the existence of a table or other entry that
tells the ADSM traffic here's where you go if your destination is
such-n-such), that ADSM traffic will indeed by routed through that specific
NIC and down the dedicated wire? I have a funny feeling that the network
folks think that ADSM will be driving when the data traffic starts
streaming toward the ADSM server, but I don't think they realize that
they've got to provide the track on which it travels first as well as the
road signs.
Thanks much!
C.L.Cannam
Storage Management
GENAM/St. Louis, MO/USA
ccannam AT genam DOT com
|
