SAN Cache Size and Cache Hit Percentage?

  • Thread starter brother_of_karamazov
  • Start date

brother_of_karamazov

Legendary Poster
Can some of you folks share your SAN cache size and cache hit percentage numbers for your production database and tempdb database?


Thanks
 
Yo Jeff

Heres a nickle. Go buy yourself an EVA . . .

grin.gif
grin.gif
 
My company has had numerous problems with the HP eva..to the point that we are implmenting other solutions, quick..didn't know if others experiences with HP was the same or maybe our volume is just too much for them.

I am not in the storage teams, so if you ask 'what problems' or 'what solutions we are moving to' I would have to ask. All I know is that the HP went down a number of times (bringing down most of the company) and that the trust eroded enough to persue other solutions..

-John
 
[ QUOTE ]
Yo Jeff

Heres a nickle. Go buy yourself an EVA . . .

grin.gif
grin.gif


[/ QUOTE ]


LOL! I've already had the numerous arguments about virtualized SAN's. RAID 5 for everything, let the software handle it.
 
...or even better, an XP. We're installing an XP12000. We have a couple of EVA's, but they're considered slow Tier 2 storage by our storage admin group. Elitist snobs.
 
XP's have had notorious issues with JDE customers. I know at least 2 customers that won lawsuits against HP over the XP. We're talking 7 digit lawsuits here.....

XP's are massively expensive - and very different from the EVA's. They come from the old mid-range enterprise systems side of HP, whereas the EVA's come from the Compaq Server (Proliant) division. When I asked about the major differences between EVA and XP's - HP actually replied that the EVA was faster, but the XP was more expensive. Figgerthatout !

You can pick up a fully stacked EVA5000 on ebay for $40k now - WOW ! like 4Tb for $40k and 10Tb for like $80k - AND HP will continue the maintenance on them (usually they're off lease equipment that HP's distributors stick up on ebay).

I just need 3 phase power in my office to run one....

Anyway - I'll get you some stats over the next coupla days Jeff...
 
Hit rates would typically be very low for the former and very high for the latter from my experience. Which is very logical, BTW. Is this what you are observing as well?
 
Okay that's it.

Forget the EVA, the XP and the whatever.

If you want a real SAN get an IBM DS4000 series.

Actually on a serious note I haven't had any issues on the IBM SAN's either with servers connected via HBA or an entire Blade Centre.

Colin
 
[ QUOTE ]
Okay that's it.

Forget the EVA, the XP and the whatever.

If you want a real SAN get an IBM DS4000 series.

Actually on a serious note I haven't had any issues on the IBM SAN's either with servers connected via HBA or an entire Blade Centre.

Colin

[/ QUOTE ]

Funny you mention the DS4000 series....


You got some numbers for cache hit percentages? Also, what's the cache size to give the numbers some perspective?


Thanks
 
[ QUOTE ]
LOL! I've already had the numerous arguments about virtualized SAN's. RAID 5 for everything, let the software handle it.

[/ QUOTE ]

Jeff - the arguement about RAID 5 for everything is waste. If you have a gazillion servers, each with their own local storage, you will have gobs of unused disk space. (don't you love my mastery of high tech terms?) You will also use a lot more real estate in your server room, require more electicity and cooling. We did a six sigma project here and found a crazy ridiculus amount of wasted disk space. They also found a lot of underutilized servers, so they implimented a vm-ware project and a SAN project to consolidate. It did not cut down on the number of servers to support, but it did cut down on the number of physical boxes to support and we now use diskspace much more efficiently. Across the two primary datacenters in NY and Connecticut, we have hundreds of terabytes of data to manage. We had substantially more disk space before the initative.
 
Just to show I wasn't exaggerating :

http://cgi.ebay.com/EVA-5000-W-168X-36-4G-2X-2-16-SWITCH-MGT-APPLIANCE_W0QQitemZ150106930746QQcategoryZ80218QQssPageNameZWDVWQQrdZ1QQcmdZViewItem

thats 4.8Tb of space for $48,000 - and as you'll see, its able to get full maintenance from HP.

When I talk about an EVA - Enterprise Virtual Array - its about performance, not necessarily just manageability. The EVA5000 from HP can perform 50,000 operations per second per SWITCH - so a dual switch EVA 5000 provides 100,000 operations/second. Pretty darn nifty.

I am still a purist. I believe in RAID 10 over RAID 5 for Production - I don't care what JDE says in their little benchmark whitepapers (they don't run the software the same way as customers - and most customers, last time I looked, use some sort of external reporting tool to really interrogate the data !). I also believe in many arms over larger size drives - HP has tried to argue the fact that an EVA "doesn't matter" - but I disagree. I see better performance when I load up more drives, end of story.

What I love about the EVA is that your partitions are squished across EVERY SINGLE DRIVE on the array. If you have 168 drives and you carve out a 5 drive RAID 5 partition - your partition will physically exist on all 168 drives still - so you get the performance no matter what. Yeah, snapshots and snapclones are kind of cool - but most Intel shops are starving for database performance.

ok - I'll try and get you the hard info tonight.
 
I agree with you on the "more drives" approach. We've found, that with JDE on 150GB drives versus 75GB drives, the 75GB drives actually perform well enough to allow the business to function. With 150GB drives (meaning less drives) performance was abysmal.

The EVA is a fine SAN, but manageability is also important. Business Copy works *much* better on an XP. We've used the XP line for many years, from the 256 and 512 to the 1024 and now moving from XP1024 to XP12000.

I'd be interested in the reason for the lawsuit(s). If they didn't settle out of court, it would be a publicly accessible document, no?
 
Charles

I'll share with you info on the lawsuits if you share info on when you used an EVA with an EnterpriseOne Database. If you're just getting fed info from HP/Hitachi reps - then you're being misled. Business Copy doesn't exist on the HP EVA - instead, they have Snapshot/Snapclone which is instantaneous. Its very, very cool - and it works brilliantly given the $100,000 pricetag that the EVA sells for from HP.

I'm sure that for certain large unix implementations, the XP series is very valid - I've helped implement a few in my time (including recommending the architecture for your company) - but I'd expect the XP12000 costs in excess of $2m. You can't SPEND that much on an EVA !

They're marketed by the same company - HP - and they're marketed to different computer models. Certainly the EVA is the fastest Microsoft SQL Server drive array that you can buy today - given that its actually faster than the XP1024 gives you a really good clue to how well the device works.

What I want to state though is that many companies get confused over what is a SAN and what is a NAS. The EVA and XP are Fibre-connected Storage that use a networking technology to connect drives to computers. Network Attached Storage, on the other hand, is little more than a USB connected hard drive in comparison ! NAS is DEFINATELY something to stay WELL clear of when looking for Database Storage - I've heard unbelievable nightmares about companies purchasing NAS devices for their database server. Believe me that drive configuration is EXCEPTIONALLY important when it comes to design of your EnterpriseOne architecture. Make a bad choice to try and save a little money ? You're going to suffer for it.

Given the fact that OneWorld costs usually at least $500,000 for the software, let alone the million or so for the implementation costs - I am always astonished at how little money is spent on putting a good architecture in place. Quite literally - if you get it right, then it really is a hands-off, black-box-in-the-corner solution. Get it wrong, and its going to be your worst nightmare every working hour !

Just a couple c's
 
We use an EVA8000 *right now* with our DEV and QA databases, including JDE 8.9 and 8.11. They are physically separate database instances from our production instance, which is on an XP1024, migrating to an XP12000.

Basically, the EVA doesn't provide the reliability and uptime requirements of the business. SLA's weren't met when EVA's were used for production...and I'm talking about large file and print servers, not JDE.

[ QUOTE ]
Business Copy doesn't exist on the HP EVA

[/ QUOTE ]

Someone better tell the marketing guys at HP:

http://h18006.www1.hp.com/products/storage/software/bizcopyeva/index.html?jumpid=reg_R1002_USEN


As for my company - actually that would be the old company which no longer exists. They were rolled up into the parent company and it became another technology alongside EVA and EMC SAN's.
 
You're right. Its an option. I forgot they had that.

However, the bottom line is that 98% of customers run < 500 concurrent users, so certainly don't need an XP. However, a lot of these companies plump money down for really bad disk arrays - especially the SQL Server companies. The EVA is ideal for these customers, far more so than the XP or the IBM Series in my opinion.!

If you're using the EVA 8000 in your company, then someone there also believes the EVA is the best choice for that requirement !
 
[ QUOTE ]
You're right. Its an option. I forgot they had that.

However, the bottom line is that 98% of customers run < 500 concurrent users, so certainly don't need an XP. However, a lot of these companies plump money down for really bad disk arrays - especially the SQL Server companies. The EVA is ideal for these customers, far more so than the XP or the IBM Series in my opinion.!

If you're using the EVA 8000 in your company, then someone there also believes the EVA is the best choice for that requirement !

[/ QUOTE ]

If I could get companies to configure what I call "virtualized" or "managed" SAN's in RAID 10 I would love it. However, most of the time they want to do one big RAID 5 and carve it up virtually from there. I just don't see it having the performance of a "virtualized" RAID 10 or a properly partitioned SAN like the IBM DS series where I can put the important dB components (tempdb, tlogs, data, etc.) on separate sets of physical spindles.


Thoughts?
 
[ QUOTE ]
Can some of you folks share your SAN cache size and cache hit percentage numbers for your production database and tempdb database?


Thanks

[/ QUOTE ]


So all this and I still didn't get any numbers?

:)
 
During our Xe to 8.9 upgrade, we pushed the storage group (was the UNIX group at the time) to implement RAID 10 for certain aspects of the database. It helped us meet our target goal of no more than 36 hours for the entire table conversion process. With RAID-5 we were looking at more than 50-60 hours (I can't recall if it was the larger of the two numbers).
 
Based on your question, this is the reply I received from a very savvy SAN admin. You mentioned an IBM SAN in another post - is this what you're dealing with?

"I would need a lot more information to be able to answer this correctly. In simple terms there is 8 gig of cache in an EVA 8000 as opposed to 32 in the XP1024 and 48 in the XP1200. Cache hit ratio’s have less to do with RAID configuration and vastly more to do with IO stream such as Random vs. Sequential and read vs. write. Raid configuration and Raid 5 concerns are vastly different between each of the 3 arrays listed. I would need to know what the concern was and which array was of concern."
 
Back
Top