SUMMARY: NetBackup 3.41 Server Sizing

A big thanks to everyone who replied (in some cases, repeatedly after
very usefull dialog :-). [Al Hopper, Travis Freeland, Adrian Nardi,
Kevin Buterbaugh, "Nadine" - in no particular order  ]

My query (see below) was regarding real-world "capacity planning" of
a Netbackup server (CPU, Memory, IO) and also the possible
feasibility of having a light-use fileserver (nfs, samba) on the same
box (for me, putatively a Blade 100 with ~ 1 gig ram, internal mirrored
IDE disk for OS slices and external SCSI disk brick for NFS,
samba shares. )


The general concensus to my query was,

-Netbackup server will be first and foremost constrained by bandwidth
issues & IO wait, not by CPU (so long as you are *not* doing data
compression in software on the NBU server).  Thus, having Gigabit ether
would be more important than CPU in order to be able to stream enough
data if using multiple drives simultaneously, or if using
higher-bandwidth drives such as "super-DLT" or LTO.

-in order for NBU to run "happily" it was recommend to have Ram
available on the system "approximately" as follows:  ~128 mb for
netbackup software, ~ 128 meg per tape drive attached to the NBU
server.  Disk space of ~150 megs is fine for a vanilla install, but be
wary of gradually growing backup database files (ie, tape backup
indices) - their size with depend ultimately on expiry of backups
[not-too-huge if expiry < 3 months .. getting quite large if window is
longer than this..].

-General recommendations were made regarding "optimizing NBU" as
follows:

    --use as many drives simultaneously as your network bandwidth will
support (ie, simultaneous streaming / multiplexing of data).  Then, buy
"this many more drives" to accommodate duplication of tapes (for
generating off-site replicates ..)

    --good NBU server planning includes good scaling capacity
(Ram, I/O, CPU). However, test! in your environment to avoid
"overbuying" for your NBU server.


-Specific recommendations were made regarding hardware:  A blade 1000
would probably be a better choice as a "longer term solution", having
the option for 2 CPUs (and faster ones at that).


My secondary query, [regarding - NFS / Samba Fileserver on
?same? blade100, capacity issues?] which was somewhat buried in the
original posting, was not addressed by any replies except one, which
commented:

-a blade 1000 would be a better choice in the long term, or failing
that, a Solaris-X86 box makes a very nice fileserver with plenty of
horsepower on a budget. [ie, Tyan Tiger MPX motherboard with
Dual-Athlon XP/MP cpu's is a fairly zippy box on a budget
that works with Solaris 8x86]

-fileserver (nfs, samba) is again limited first and foremost by I/O, not

CPU.  *Certainly* be sure to avoid having exports hosted on IDE
drives, since IDE disks have poor "concurrent user performance"
compared to SCSI.

Overall, I get the impression that people have used machines comparable
(or somewhat smaller) for larger NBU backup environments than I'm in
currently, so I'm quite optimistic that the blade 100 will be a
perfectly reasonable "not-long-term" solution. [ Same seems to hold true

for FileServer functionailty, so we'll see how it all works out.]

Many thanks to everyone who replied,


--Tim Chipman






=====================================================
ORIGINAL QUERY:


Hi Folks,

Sorry if this is terribly obvious, but google searches and archives for
this list don't appear to have much on this, alas.

I'm curious if anyone can comment on the "real" requirements
(especially WRT CPU horsepower) required to operate a Netbackup
"server"  (ie, the machine to which the tape library is connected via
Scsi, and which connects to netbackup client  machines  [both servers
and workstations] via TCP to do backups.

The veritas website only mentions "128 megs ram (256 recommended) plus
~150 megs disk space" as the requirements for the NBU server.

I am currently suspecting that it is more of an IO-bound kind of
function, rather than a CPU-bound issue (we're not using additional
compression of data streams!), but wanted to check with people in this
list who might have real-life experience with this sort of madness :-)

For instance: Could a Sunblade 100 (500mhz Ultra2e CPU) be up to the
task of acting as a netbackup server box, assuming we have a decent
amount of ram in there (between 512-1024 megs ram probably) ? (It would
be using 100 mbit ether for the moment I suspect)

All this is brought up in my mind because I've got one e450 box
currently wearing "three hats" - filling the role of an Oracle DB server

; a Netbackup server (attached to an L1000 robot with 2 DLT7000 drives,
connected to the network via 100mbit ethernet) ; and also a fileserver
(NFS, Samba plus Totalnet for a few mac clients - serving files from
an A1000 disk array.).  I've noted that with our new backup schedule,
iowait loading is very significant when the backups are running, and I
think oracle in particular is suffering. Likewise, if the fileserver is
getting hammered with disk access, then oracle is suffering as well.

My "temporary cunning plan" was initially to move just the fileserver
functionality over to a blade100 - I figure this is as much horsepower
as an Ultra5, which should (?) be plenty to drive a low-to-moderate
volume fileserver.  However, after a bit more thought, I realized that
moving netbackup might be a big  improvement to this picture as well.
Assuming the blade 100 is doing *nothing* other than fileserver (approx
a dozen NFS clients, none "heavy", and ~max 20 workstation users via
samba) .. it seems that all this functionality is IO bottleneck, so the
moderate "low end" scale of a blade ... ?should? be enough to power it
[of course, I'd be putting into this blade unit two seperate Diff-Scsi
cards, one for an A1000 disk brick and the second for the L-1000 -- so
in theory the internal IDE disk performance should not be a big factor
to this (?).

[ I realize that a blade 100 is *not* a "super redundant server" (ie,
dual power supplies, etc) - but at very least, I can get disksuite
mirroring on a pair of internal IDE OS disks to provide fault tolerance
there, I guess.. ]

Anyhow. That's my big wordy query. In short, it might read:

"any comments on real world requirements for a combo netback server,
NFS/Samba Fileserver box" ?

As always, I'll summarize and post the results once people finish
peppering me with advice :-)

Thanks muchly,


--Tim Chipman
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers