|
Re: [Networker] Sun Hardware
2009-10-14 05:58:33
Joel Fisher wrote:
Hey Guys,
Can anyone give me an idea of how many lto-4 drives a T2000 or T5220
could keep streaming.
You cannot really compare the two, because they use different
processors. It also depends on how many cores you buy and on how many
sessions you run. Since every drive is served by its own nsrmmd, then it
is very likely that each core will be driving a drive. Since these
processors are also 4-way multi-threaded, I suspect that you might even
be able to drive (CPU wise) two or even three drives with each core.
Specifically, my backup server is a 6-core 1Ghz T1000 and every
core/nsrmmd can do ~100MBps if it doesn't need to multiplex too many
save sessions. I usually run with ~30 sessions per drive so nsrmmd
becomes busy and runs at ~40MBps. If it gets ~10 sessions, speed can get
to over 70MBps. I have four LTO-3 drives.
Assumptions.
Data is landed on disk(adv_file) then cloned to LTO-4 drives during the
day.
The disk is not a bottle neck.
This statement is not trivial. I have used UFS in the past and moved
to ZFS and there is a huge difference between the two. Also, with ZFS I
used two setups. I have an AX150 with 12 SATA disks. At the beginning I
had 500Gb disks which I presented to ZFS as a single LUN over a single
RAID5 group which was 10D+1P+1H. Then I moved to 750GB disks which I
presented to ZFS as three LUNs each over its own RAID5 group with 3D+1P.
The second setup was faster. EMC's best practice guide recommends RAID3
with 4D+1P, but the AX150 doesn't do RAID3.
Plenty of hbas so connectivity is not a bottleneck
I also expect you to get yourself a 10GE card.
Could I have say 3 landing areas and then stream to 3 LTO-4
simultaneously. Somewhere between 120-240MB/s?
With these setup your staging can benefit from six cores, because
staging is using two nsrmmd processes, one which reads from the RO
instance of the aftd and one to write to the drive. Unfortunately, it
seems that Sun don't deliver the 6-core version any more and anyhow you
would probably want the faster cores (they have 1.2Gh, 1.4Ghz and
1.6Ghz). Checking Sun's web site you will see that he problem is that if
you go for the 1.6Ghz core, you end up with a 45K$ T5220 machine, 8
1.6Ghz cores and 64Gb RAM which is an overkill. Or, you can hope that
the 1.2Ghz core will do and settle for a 15K$ T5220 machine, 4 1.2Ghz
cores and 8Gb RAM. If you go with a T2000, you can get a 21K$ machine, 8
1.2Ghz cores and 16Gb RAM. Keep in mind that HBAs and 10GE card will get
you to an even higher price point. However, since you are probably
entitled for an educational discount, you should probably speak to your
Sun's account manager and check what you can get (maybe they will have
other configurations as well).
Any real world users of something like this.
I know these servers can seriously move some data, I guess my main
concern, is can they move a single stream fast enough for LTO-4 or
LTO-5? Since any individual thread is limited to the 1.2-1.6Mhz.
Extrapolating from my data, I can guess that the 1.6Ghz core will be
able to feed an LTO-4 drive at native speed (120MBps). I am quite sure
it will have problems with feeding a compressing drive. Feeding an LTO-5
drive (180MBps) seems like a real issue. Also, my understanding is that
if you are not feeding the drive fast enough, it might not bother with
compression at all which means that you also get lower capacity from
each cartridge (I am sure I read it somewhere, but cannot verify it
right now, so I suggest you will re-check this urban myth).
Any advice here would be appreciated.
Thanks!
Joel
To sign off this list, send email to listserv AT listserv.temple DOT edu and type
"signoff networker" in the body of the email. Please write to networker-request
AT listserv.temple DOT edu if you have any problems with this list. You can access the
archives at http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
--
-- Yaron.
To sign off this list, send email to listserv AT listserv.temple DOT edu and type
"signoff networker" in the body of the email. Please write to networker-request
AT listserv.temple DOT edu if you have any problems with this list. You can access the
archives at http://listserv.temple.edu/archives/networker.html or
via RSS at http://listserv.temple.edu/cgi-bin/wa?RSS&L=NETWORKER
|
|
|
