Bill, yes, MPEiX works better with many spindles on the existing drives
AND memory, but with MPEiX 6.5 you will be able to almost double the
memory installed in many of the boxes. If you increase main memory well
beyond the normal allocations for users and process, i.e. 3GB or 4GB
systems instead of 1GB or 2GB systems, you don't see the spindle impact.
You can also improve by not placing all 10 drives in the HVD10 on one
FWSCSI channel. I'd also strongly recommend anybody using SCSI drives
in either the HAAS or HVD10 to invest in disk mirroring.
We lost another FWSCSI drive on a machine this week, the current score
for drives is something like 12 and 3, i.e. 12 failures and 3 times we
were warned by predictive detecting possible failures.
Bill Lancaster wrote:
>
> Hi,
>
> I'd like to add another caution to the use of the HVD10 disks. They come
> with 9, 18, 36 or 72 gb disks, with 10 to an enclosure. This means that
> you can get up to 720 gb in the space roughly that of a Jamaica.
>
> As most of you know, one of the main majiks about MPE is how tightly
> integrated the whole package is. Hardware, software, operating system,
> file system and DBMS (Image) all play very well together (btw, this is true
> of other "proprietary" systems as well such as the AS400). This being the
> case, there is a very real, very tangible performance advantage. Those of
> you who have Unix/Oracle (poor, poor souls) experience the relative
> differences in performance every day. I've often heard (maybe because I've
> often said :-) that, if one was to take an HP 3000 application running
> VPLUS and Image and ported it to Unix under Oracle, it would take
> approximately four to eight times the amount of horsepower to function
> comparably to the HP 3000 environment.
>
> One of the reasons this is true is because of the wonderful I/O efficiency
> the HP 3000 offers. For example, one of the best I/O performance metrics
> on the 3000 is Read Hit Percentage. This is the number of read requests
> satisfied in main memory. The other side of the coin is that this
> indicates what percentage of all physical disk reads are eliminated. For a
> healthy HP 3000 this number is *generally* 95 percent or greater. (For the
> benefit of my friend James Reynolds let me add an "It Depends" here.)
>
> Given the fact that nearly all physical disk reads are eliminated and that
> nearly all disk writes are journaled through the Transaction Manager, most
> I/O performance problems are removed from the response time of the
> individual user. (As an aside, this is why a health MPE disk environment
> won't benefit much from the new whiz-bang disk I/O technologies. The most
> additional physical disk I/O that can be eliminated is in the 5-10 percent
> range, not an amount likely to be visible to the end user).
>
> One of the reasons disk I/O is so efficient on MPE is that there is a
> significant amount of disk I/O parallelism. That is, the more spindles you
> have, the more each spindle can be concurrently tasked with I/O
> requests. Given that the most heavily accessed files on any particular
> system are likely broken into multiple pieces (or "extents") and that these
> pieces are likely located on multiple spindles, the I/O subsystem will
> generally be multitasking on your behalf. Add to this the inherent
> additional multitasking associated with Mirrored Disk/iX (reads are
> "balanced" on both of the mirrored pairs) and you have a wonderfully
> efficient I/O environment.
>
> This leads me to the essence of this posting. The driving technical issues
> behind the newer disk technologies do not take into account the so-called
> proprietary environments. IMO the three main driving factors behind these
> new technologies are 1) Price, 2) Density and 3) High Availability, not
> necessarily in that order. I believe that these factors are driven largely
> by the Unix and Wintel markets, the lowest common denominators of computing
> today.
>
> The price and density factors essentially result in fewer and fewer
> spindles to spread out your data over. Given that I/O performance in a
> Unix/Oracle world generally, by our standards anyway, sucks (a technical
> term) you don't generally perceive such a loss of spindles in that
> environment. (Before some of you Unix heads get upset with me let me add
> that this isn't always true. Just most of the time. :-). In MPE-land you
> experience it pretty dramatically.
>
> That all being said, as you make the sojourn into the nether region of
> Unix- and PC-led disk hardware, please, please, please keep in mind that
> you have to "care-and-feed" an MPE environments performance
> differently. You must have a care to ensure that you never dramatically
> drop the number of spindles you spread your mission-critical data over. We
> are in the process of implementing and HVD10 disk solution for a
> customer. From the beginning we are 1) selecting only the 9gb disks (fast
> little critters!) and 2) configuring only 50 percent of the space. This
> allows us to actually have a performance advantage of the customers
> existing disk environment by not measurable reducing the spindles and by
> having much faster disks.
>
> This may seem wasteful but there isn't much price difference between the
> Jamaica 4.3's and the HVD10 9.0 gb disks.
>
> Hope this helps.
>
> Bill Lancaster
>
> At 09:28 AM 08/08/2000, John Painter wrote:
> >Jim:
> >
> >I checked this out on your heads up-- took me by surprise too. The short
> >of it is that the RACK mount HASS enclosures are being obsoleted in
> >favor of this new HVD10 product that supports faster and bigger disks.
> >The DESKSIDE HASS will remain until some future product TBA in 2001
> >takes it's place. The HVD10 product is more expensive to buy, but it
> >holds more capacity in it's footprint. The kicker is that you need to be
> >careful which disks you order under MPE, because some of the disks
> >supported in the enclosure are not supported under MPE. Meanwhile, HASS
> >parts (disks, fans, power supplies) will be available till mid 2001 and
> >there is the standard 5 year support life. There is also the typical HP
> >trade in program to move your rack mount HASS out and the HVD10 in.
> >
> >Hey, what have you bought from HP lately? ;-)
> >
> >John Painter
> >Computer Solutions, Inc.
> >http://www.internetcsi.com
> >
> >
> >
> >
> >Jim Phillips wrote:
> > >
> > > I read where:
> > >
> > > "High Availability Storage System (HASS) (A3312A/AZ) Rack Units will be
> > > Discontinuance on November 1, 2000. This will be the 6-month notice. The
> > > customers are recommend to go to the HP SureStore HVD10 product
> > (A5616A/AZ).
> > > Please see the ESP and CNO announcement on the HVD10."
> > >
> > > Just what does this mean?
> > >
> > > Jim Phillips Manager of Information Systems
> > > E-Mail: [log in to unmask] Therm-O-Link, Inc.
> > > Phone: (330) 527-2124 P. O. Box 285
> > > Fax: (330) 527-2123 10513 Freedom Street
> > > Web: http://www.tolwire.com Garrettsville, Ohio 44231
|
