RE: 11.50.xC4 Compression with workgroup edition?

> Date: Fri, 22 May 2009 19:59:32 +0100 > From: obnoxio@serendipita.com > To: informix-list@iiug.org > Subject: Re: 11.50.xC4 Compression with workgroup edition? > > Ian Michael Gumby wrote: > > Blah, blah, blah, blah ... > > > So maybe, just MAYBE, IBM isn't targeting the SMB user with this feature? > > But hey, what do I know? It's not like I'm not perving over pictures of > Lindsay Lohan in a three-way with Britney and Sharon Stone! > You are truly a sick man. Ok, so if we look at IBM's newly released SSDs http://www.theregister.co.uk/2009/05/22/ibm_power_ssds/ A single 120GB SSD formatted yields 69GB at a cost of $10,000.00 Now that for a very fast system that can give you a lot of I/Os per second. IBM also formats the drive to counter the issue of wear that occurs in SSDs. So a 4 Disk Raid 10 would list out at $40K and yield roughly 138GB of storage space. If I look at the cost of compression on an 8 core Xeon box, the cost is 120K just for the feature. Suppose that there's a 50% compression on my tables. (Thats a big if). Adding 4 more SSDs would only cost an additional 40K. So even for the 'big boys, think about adding disks is probably a better deal. But this is at retail for both the hardware and software. But then again, the big boys don't pay retail or even close. How much of a discount is J level? And then if you are a big enough player to get J level discounts, you can still press your reps to give you even a percentage off of that too. So yeah, at a *75%* discount from list... *maybe* this feature does make sense. ;-) But how many customers can get that from IBM? -G _________________________________________________________________ Hotmail® goes with you. http://windowslive.com/Tutorial/Hotmail/Mobile?ocid=TXT_TAGLM_WL_HM_Tutorial_Mobile1_052009

Uhm... So how often can you update any given row on an SSD drive? before it burns out? SSD in the warehouse yes...SSD in OLTP... not so much - yet. OTC is right about the target market. Cheers Serge -- Serge Rielau SQL Architect DB2 for LUW IBM Toronto Lab

Serge Rielau schrieb: > Uhm... So how often can you update any given row on an SSD drive? before > it burns out? SSD in the warehouse yes...SSD in OLTP... not so much - yet. > OTC is right about the target market. > > Cheers > Serge Hi Serge, sry, but I cannot second this. We now testing NAND flash SSDs from Intel and other manufacturers. All are in the EUR 600-700 price range and have 256GB each. hey are either MLC or a hybrid of SLC & MLC flash memory. (239 formatted) We run those now for more than 3 months, only writing at an avg of 2000 write operations per second. We want to see write amplification and other ugly things like getting rotten or slowing down. So far no problems. Given that I run our databases with a write cache rate better than 80% 2000 write OPS for more 90 x 24 hrs is MUCH more than a disk would have to do in 2 years. At a price tag of less than 1000 USD, I'd have no problem to proactively change the disks for new ones after 12 months. www.storagesearch.com has nice articles & white papers about all this SSD business. TMS is in lead for long time now when it comes to SSD subsystems which are really fast (and epensive). Those are sold to customers in need of more than 500K IOPS. Toshiba recently announced the poor man's MLC flash SSD having 512GB and will sit in your laptop. Get a gamer laptop with 3 slots and you are set. This laptop can offer ~ 7500 reads per second AND 4000 writes per second at a street price of less than 4K USD. It will offer 8GB of RAM for beeing used as a NFS server with lots of cache. Use eSATA connection (a good laptop has 2 eSATA connectors) to bring speed to your big iron ;) I tend to tell my customers that if they want more performance they must start to spend less money. dic_k -- Richard Kofler SOLID STATE EDV Dienstleistungen GmbH Vienna/Austria/Europe

Richard Kofler wrote: > Serge Rielau schrieb: >> Uhm... So how often can you update any given row on an SSD drive? >> before it burns out? SSD in the warehouse yes...SSD in OLTP... not so >> much - yet. >> OTC is right about the target market. >> >> Cheers >> Serge > > Hi Serge, > > sry, but I cannot second this. > We now testing NAND flash SSDs from Intel and other manufacturers. > All are in the EUR 600-700 price range and have 256GB each. > hey are either MLC or a hybrid of SLC & MLC flash memory. > (239 formatted) > We run those now for more than 3 months, only writing at an avg of > 2000 write operations per second. We want to see write amplification and > other ugly things like getting rotten or slowing down. > So far no problems. > Given that I run our databases with a write cache rate better than 80% > 2000 write OPS for more 90 x 24 hrs is MUCH more than a disk would have > to do in 2 years. At a price tag of less than 1000 USD, I'd have no > problem to proactively change the disks for new ones after 12 months. > > www.storagesearch.com has nice articles & white papers about all > this SSD business. > > TMS is in lead for long time now when it comes to SSD subsystems > which are really fast (and epensive). Those are sold to customers > in need of more than 500K IOPS. > > Toshiba recently announced the poor man's MLC flash SSD having 512GB and > will > sit in your laptop. Get a gamer laptop with 3 slots and you are set. > This laptop can offer ~ 7500 reads per second AND 4000 writes per > second at a street price of less than 4K USD. It will offer 8GB of RAM > for beeing used as a NFS server with lots of cache. Use eSATA connection > (a good laptop has 2 eSATA connectors) to bring speed to your big iron ;) > > I tend to tell my customers that if they want more performance > they must start to spend less money. > > dic_k SSD and logical log storage... The writes are perfectly split across all of the drive... And it's easy to calculate the rate of write operations per day/hour/second/whatever. And logical logs can be a major performance challenge... I'd suggest you start there :) As for Gumby's arguments... Thos kind of systems are very far from the ones used by many customers (big shops). And you're looking at the price of acquisition. That's precisely why I wrote at the beginning that I have great doubts that many customers are able to do the maths... Many of them tend to do the same kind of calculations. You're forgetting about all the rest... "second line" instances, backups, time to do certain operations etc.... Regards. -- Fernando Nunes Portugal http://informix-technology.blogspot.com My email works... but I don't check it frequently...

On May 22, 4:10 pm, Serge Rielau <srie...@ca.ibm.com> wrote: > Uhm... So how often can you update any given row on an SSD drive? before > it burns out? SSD in the warehouse yes...SSD in OLTP... not so much - yet. > OTC is right about the target market. > > Cheers > Serge Well Serge, perhaps if you stepped outside of the really nice IBM office in Markham, you might understand what takes place in today's market. Perhaps I have a better view of the market from my building where I have views of the John Hancock, The Trump Building, and the Sears Tower. (And of course all of the buildings in between.) The views from Markham? Not so much. ;-) Since you asked about MTBF, depending on your source and how they did they calculations, MTBF varies from 1.4 million hours to 3.9 million hours. This of course depends on the underlying technology SLC vs MLC and how the drives were tested. According to other google found sources, 2.5" SAS drives have a MTBF in the 1.5 million hour range. Now with respect as to the relative value of compression when used in an OLTP environment vs a DW environment... If the data lookup is frequent and speed is of the essence, most likely the application/app server will cache the data in memory rather than in the database. If you need to gain fast access of the data, and you're in the database, then you have to consider that you'll want to try and hit the index only rather than try and hit the underlying table. Again here, you don't gain anything in terms of performance with compression. With respect to a DW, you still have to consider cost of the compression feature versus alternatives. Using our test system of 8 cores of Xeon 5500 series chips, 120K buys a lot of alternative disk/memory/etc hardware. I'm sorry but I can't see customers, even big customers, shelling out $ $$$$ for this feature. Of course depending on the big customer's discount level (J level - 90%?) this feature might make sense. Oh and Serge, I don't know if you actually read anything that isn't hype from IBM, but if you can get your hands on a recent copy of CRN, thats Computer Reseller News, you'll see that they took a system, and created 18 or so virtualized Microsoft Servers. I don't have the issue in front of me and I don't recall the specifics. The point was that with today's hardware, using virtualization can cut down on the size and expense of your machine room. So this little box isn't just for the SMB market. In fact, if you consider virtualization, Microsoft SQLServer, this little box would be perfect for some of IBM's soon to be ex-customers as they migrate from DB2 on z/OS to SQLServer. (Two of these customers happen to be in Chicago. ;-) You could probably run Hyatt's centralized reservation system on this box, but that's just a swag on my part. But hey! What do I know? I never drank the blue cool aid. ;-) -G Of course if IBM ever did a benchmark and released it of IDS on this type of server... well we can see what a *real* database can do. But that will never happen. We wouldn't want to embarrass DB2, now would we? ;-P