I'll take one more pass before going offline. I'm feeling rather Wirt-ian
today so if this topic isn't of interest, delete now!
At 12:20 PM 3/18/98 -0600, David Lethe wrote:
>I respect your opinion, Bill, but I feel you don’t have all the facts:
Actually, I never said I had all the facts. I certainly don't have all the
facts. I'm just advocating caution, not elimination.
>
>You wrote it is not “officially” supported by HP. Duh!! How many
third-party
^^^^^ ouch!
>disk subsystems does HP support? It is political. Here we have a disk
>subsystem that can potentially double or triple the workload, and prevent the
>customer from doing an unnecessary CPU upgrade. It also provides other
>advantages over JBOD and EMC.
>
I know for a fact that HP spent enormous energy in certifying EMC for
resale and support into their customer base. As to David's assertions
about doubling or tripling the workload, I'm not denying the possibility.
Rather, I am saying that the performance implications are very interesting
to me. I do take issue with his implication that doubling or tripling the
workload will result in the customer avoiding an "unnecessary CPU upgrade".
First, I find it mildly offensive that he would imply that HP is more
interested in selling the system upgrade than in resolving I/O bottlenecks.
I see a tremendous concern about that very issue within the halls of CSY.
Second, no matter how fast disk subsystems are they are never going to
improve the performance of a saturated backplane. Third, it doesn't
necessarily follow that the knowledgable customer would attempt to purchase
a CPU upgrade to address a disk I/O problem. If there is a latent CPU
shortage hidden by an disk I/O bottleneck, increasing the disk I/O
throughput will only put more pressure on the CPU anyway, thereby
necessitating a more immediate CPU upgrade. (BTW, I'm not saying that some
customers don't purchase CPU upgrades to address a problem with disk I/O.
Continuing to educate customers on these issues remains an important issue,
one I, and others, attempt to address at client sites, users groups,
hp3000-l and so on). Also, I get hives when people simplify performance
issues like this.
>>At the very least, David's company should get HP to say SSA does no harm
>>and is at least as robust as JBOD (Just a Bunch Of Disk).
>
>HOW WE HAVE TESTED SSA TO DATE:
< really good stuff snipped>
These test issues are all great. Again, I am for any reasonable, solid,
tested solution for disk I/O issues on the HP 3000. Given the incredible
growth of disk storage requirements in the computing world, and the
likelihood that it will continue at an ever-increasing rate, new disk
technologies are going to be essential to the continued success of the HP
3000. I am just basically conservative when it comes to introducing
new-to-the-HP3000 technology without having the technology exposed to the
creative minds behind MPE. I'm not saying that David hasn't tried. I
would like to see HP step in and aggressively test this technology. My big
concern is that with the HP 3000 there exists an extremely tight
integration between the operating system and the file system. This is why
most HP 3000 sites running EMC don't get the same type of performance
improvement with EMC cache as our brethren in the Unix world get: nobody
eliminates a physical disk I/O on MPE better than MPE (for the most part).
Prefetching is a deeply ingrained activity within nearly all types of MPE
disk I/O. And it is sized in accordance with the physical hardware
architecture.
>features, and more redundancy). There are also some additional enhancements
>coming soon which will knock your socks off … such as getting beyond the
>255 disk drive limit!
Now I'm really concerned. The 255 spindle limit isn't trivially "fixed"
within MPE (not that I know any details about how to change it). Before
anybody getting too excited about this feature, it needs to be CLOSELY
examined by people very familiar with MPE's file system.
In summary, let me say the following:
1. I am very much in favor of any solid, reasonable, reliable innovative
disk I/O technology for the HP 3000.
2. Customers should take a conservative approach with new disk
technology, not because I make money on HP disk sales (I don't :-( ) but
because of supportability, track record and "bleeding edge" issues.
3. Anybody who knows me knows that I'm not a big fan of EMC (the
company), the performance of their products, the pricing of their products
or their performance claims but there is a set of conditions under which it
is appropriate to use their products. My hope is that with SSA this set of
conditions increases but I will not recommend a third-party disk technology
product which hasn't been vetted by HP-CSY. The product doesn't have to be
RECOMMENDED by HP, just vetted. This is just my opinion but knowing the
competitive pressures HP is in when it actually tries to sell EMC
solutions, HP would probably embrace a technology which offers better
price/performance and a higher degree of accessibility to customers. After
all, how many HP 3000 customers can actually afford EMC solutions? The top
1-2 percent???
4. Remember, what I advocated was "a little caution" not "throw the baby
out with the bath water".
I would challenge HP to step up to the plate, test this technology and
report back their findings. I would challenge David to provide us with a
list of HP 3000 customers currently using this technology in production. I
would be happy to volunteer my services to examine the *actual* performance
of these sites and report my findings back to this group (with their
permission, of course). Let's have a shoot-out with JBOD, EMC and SSA
technology! I'm happy to spend lots of time on this.
>
>Bill, how do I get in contact with this executive forum on high
performance???
>
You just did :-) Feel free to contact me directly for more information.
Unless the list wants to see continuing dialogue on this discussion, we can
take it offline from here. Listers???
Bill (Duh!!) Lancaster
Lancaster Consulting
[log in to unmask]
541-926-1542
|
