Dear Sun Managers,
with solved original problem comes this summary (original post trailing).
Not strange that this post triggered few responces. Yet it pleasantly rewarded
me with several discoveries - which are summarized together with results of
local searches:
1st - Thanks to:
Reto A Lichtensteiger rali@meitca.com
Dave McFerren davem@cp.tybrin.com
Brad Young bbyoung@amoco.com
John Lees john@ushers.demon.co.uk
a) weekend posts have low chances to be looked at on Monday - since list
assumes that problem is solved within 24 hours
(~90% true, rest 10% > /dev/null)
b) 9GB vs $.3GB -
make as many spindles as possible. While building cheap RAID look at real
RAID components - and they are usualy 2.1 GB 7200 rpm.
Unpleasant *feature* of 7200 drives - only 2 should be packed in a box
capable to hold 4 - require good cooling.
c) Micropolis vs Seagate
Micropolis LS series (9GB) are excellent - allow for very short total SCSI
chain. These are twice more expensive than just 1991 series (in Moscow).
Unpleasant 'feature' - can be very sensitive to power supply : as it
degrades to 4.8 from 5V nominal they will not spin up. Diagnostics will
be misleading - since drive will be OK power supply should be
checked and (probably) replaced
Otherwise see price tag or get a label you are used to.
d) I'd mentioned that drives will be used under ODS, stripped into single
volume:
*unrecommended* without mirroring.
Workaround for poor (well, economical):
put cron job asking metadb how replica's are feeling,
use ODS logging facility,
double backup frequency,
and still do not put vital data on that volume.
e) additional esp interfaces - one per box with 2 drives -
are highly recommended (coincides with recommendations in SUN tech.papers)
f) extra hints - upgrade to 2.5 (speculated 20-35% NFS performance gain)
- PrestoServe NVRAM/NVSIMM
-----------------------------------------
Practical Result:
2 external boxes, each with 2x4.3 GB 7200 rpm
2 fast scsi/ethernet Sbus cards
2 64MB chips
Well, it does exceed initial 7K budget, but was rubber-stamped within 15
minutes after presenting query results.
-----------a bit truncated original post--------------
Subject: PG needed: Micropolis vs. Seagate , 9g vs x4g
Dear Sun Manager,
I am seeking an advice on upgrading and configuring machine for top :-)
performance while having rather limited budget ( > ~7K $$ ).
Hardware currently considered :
64Mb RAM chip (I'd rather get 2 of them) - that goes without arguments;
1st - single external box with 2 Micropolis model 1991 9Gb disks - $4050
(to be put on existing esp);
2nd - single external box with 2 Seagate 40800N 9Gb disks - $4460
( same as above )
3rd - 2 external boxes each with single 9G (Mic/Sea - depending on vote)
plus additional fast esp card (+ $500) $5000
*** all these 9g are 5400 rpm ***
4th - 2 ext. boxes with 2 x 4Gb ST15150N (7200 rpm) + FASesp $5200
In any case drives will be stripped into single volume and run under
disksuite.
Some Parental Guidance is really required - I never touched Micropolis
drives and not sure about their reliability and speed (except ad's specs).
Any misfortunes, pitfalls and are they really nice?
on their Web page they say it can do "up to 10Mb sync". Hm, ... .
Have someone seen it to make, say, 5 MB/s in real life?
Will it be better than Seagate? Why?
------------------------------------
Current h/w, OS and usage pattern
Machine : SS20 w/ 2 60Mhz cpus , 64Mb RAM, 2 x 1G internal seagates,
2x2.1G seagates, external box, (sitting on sbus card from IPX,
stripped into single volume by ODS 3.0)
OS : S2.4, hw 3/95
used as : NFS server for 7 other Lab machines - it keeps several bio
databases on 4g volume (about 90% full by now).
web-server - running SRS (sort of homemade flatfile db manager),
several homemade number-crunchers (with up to week runtime &&
~16Mb per job, couple of them with heavy IO chewing flatfiles)
s/w planned to add:
acedb - another piece of bio db manager, with WWW interface and
with local X-frontend;
Protein Data Bank WWW software - will need ~4g disk space,
expected to be RAM && I/O intensive;
Machine is already (always had been) choking of RAM shortage, some projects
had been cancelled due to unbearable paging.
--------------------------------------------------------------------------
With best regards
Fedor Gnuchev
(hm, or Ted - in this English-typing world...)
mailto:qwe@ht.eimb.rssi.ru
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:58 CDT