Entries for the 'hardware' Category

Kicking AMD’s Dead Carcass

I wrote about the Nehalems a while back. I was about to write about AMD’s come back but they are due in 2010 while Intel will one ups them again before the end of this year. This is after, it is rumored, that they are delaying the Nehalem-EX just because they can. How is that for innovation? However, what are you going to do? An eight core four socket box with DDR3, and 96MB of cache will smoke an eight socket AMD HP  dl785 box. Things will get fun in the database area.

The Nehalem-EX Advantage

• Intel Nehalem Architecture built on Intel's unique 45nm high-k metal gate technology process
• Up to 8 cores per processor
• Up to 16 threads per processor with Intel® Hyper-threading
• Scalability up to eight sockets via Quick Path Interconnects and greater with third-party node controllers
• QuickPath Architecture with four high-bandwidth links
• 24MB of shared cache
• Integrated memory controllers
• Intel Turbo Boost Technology
• Intel scalable memory buffer and scalable memory interconnects
• Up to 9x the memory bandwidth of previous generation
• Support for up to 16 memory slots per processor socket
• Advanced RAS capabilities including MCA Recovery
• 2.3 billion transistors

email it! |   |   | 

Permalink     3 Comments

Nail in the Coffin for AMD?

Intel released the Nehalem processor family. The Intel® Xeon® Processor 5500 series family. This could be the nail in the coffin for AMD. I hope not. Without competition, Intel can rest on its laurels.

Lets look at the goodies. On Glenn Berry’s blog, he points out the SQL specific benchmarks. This comes from the Anandtech benchmarks. This is what caught my eye from that review.

The memory controller has up to three channels. A dual CPU configuration has access to 35GB/s of memory bandwidth (measured with stream) if you use DDR3-1333. The latest dual Opteron achieves 19.4GB/s with DDR2-800

Think about it. If you have a SQL box with 32GB of RAM and a VLDB, you could theoretically churn the buffer pool once every second. Of course, you will probably hit a disk bottleneck first. In addition the the proc specific improvements, DDR3 with NUMA support in a server is a huge leap.

Today, HP also introduced the DL3X0 G6. Here is a link to the the DL 360 G6 specs. I am speculating but I bet  it would beat a dual socket 6 core DL 580 G5. Most definitely on IO bound workloads like a database server. Hopefully, the DL580 G6’s are coming soon. Maybe an 8 socket DL 780 G6. :)

Mix that with VMWare ESX 4.0 that is in RC and virtualization of the database server may have come of age.

The Opteron 1up’ed Intel in 2004. Now the ball is back in AMD’s court. I am rooting for you!

email it! |   |   | 

Permalink     2 Comments

Mixed Workloads Part 2

I am presenting at the Ft. Worth SQL Server Users Group in April. Details forthcoming but it is basically going to be on running mixed workloads(OLTP and DSS) on the same server. This is part two of what I will be pulling into the presentation. Click here for part one.

In this part we will look at the typical hardware configurations used in part one(logshipping, mirroring, replication etc.). and what a better configuration would be especially if you leverage some new features in SQL Server 2008 to run mixed workloads.

Unless you invest the time to create a real reporting solution with an ETL, you end up with the following solutions based off of the methods described in part one.I see it all the time. On top of that, I see reporting queries still running on the production server because there is a need for real time data.

So what do we get with this solution? Most reporting queries are offloaded from production. However, there are a lot of con’s. The schema usually is not optimized for reporting. There is overhead in getting the data to the reporting server. The data is stored twice on disk and more importantly, memory. Finally, resource utilization is usually lopsided. For example, first thing in the morning the reporting server may be hammered while production traffic is just ramping up. During peak production traffic, the reporting server can be underutilized.

I submit to you that combing reporting and production is a better configuration if you do not invest in an ETL solution that creates a real reporting database. In the next post, we will talk about features to optimize this configuration but lets talk about what we gain just by using this architecture.

• The hardware is doubled and the amount of data is cut in half.
• In the configurations that this would replace, the memory gain is huge. Unless the reporting database is optimized for reporting, the data pages in memory are going to be the same. This allows the same page in memory to satisfy both OLTP and reporting traffic.
• This configuration gives more cores for parallel plans that are common in reporting.
• Most of all, it make available hardware that could otherwise be idle.

Note: I use direct attached storage(DAS) in these examples because that is where the biggest gains are to be had. However, the same benefits apply if you are on an enterprise level SAN with some caveats.

Note 2: This series is generalized and your mileage may vary based on your particular environment, business requirements and workload.

In part three, we will talk about features that will help optimize a mixed workload on a single instance with feature in SQL Server 2008.

email it! |   |   | 

Permalink     2 Comments

Another Overlooked Windows Setting for the DBA

*Warning* Only use if your array controller has a battery backed cache. *Warning*

The settings are “Enable write caching on the disk” and “Enable advanced performance”. You can access these through device manager on the properties of the disk. These settings mostly apply to direct attached storage and are unavailable for most enterprise SAN lun’s that I have seen.

While we are at it, if your RAID controller cache has a read\write ratio, it is a good idea to set it to 0% read\100% write as long as you do not have a memory bottleneck. SQL uses RAM as its read buffer.

So is this a silver bullet for performance? Definitely not especially if you are not hitting a disk write bottleneck. However, every little bit helps and if it knocks 5-10% off of your 3 hour long full backup to disk, that is a win!

Happy Friday!

email it! |   |   | 

Permalink     1 Comments

The SAN, as we know it, will soon be Obsolete

The high end applications for mission critical business are mostly powered by EMC. Other companies like 3PAR, NetApp and HP are biting into their market dominance. However, EMC has a bigger problem looming.

I have written about Solid State Drives(SSD) before but the Intel business class drives are the only ones that come close to being production ready due to random write speed. The other problems of price and capacity keep them out of reach for most companies. On the consumer side, the high end Seagate conventional drives beat most of the SSD drives on write performance.

The stepping stone

Tiered storage will probably be how SSD make their first appearance in production apps. Maybe SATA shelves for backups, SCSI or fiber channel drives for 80-90% of the data and SSD for the really hot data.

The game changer

FusionIO’s latest press release changed all of that. Imagine putting the capacity and performance of those 5 cabinets shown above in a shelf with 1% of the power requirements. Note: The TB+ capacity cards are coming Q2 2009. They currently support up 640GB. Here are some numbers on a 4 card setup:

Here are single card numbers:

• Sustained read bandwidth: 1500 MB/sec (32k packet size)
• Sustained write bandwidth: 1400 MB/sec (32k packet size)
• Read IOPS: 186,000 (4k packet size)
• Write IOPS: 167,000 (4k packet size)
• Latency < 50 µsec

What is missing?

A chassis. Sure, you could put 4 or 6 of these in a server depending on how many slots you have but you are going to saturate the PCI bus before they are fully utilized. Before that happens, you will probably hit a CPU or memory bottleneck. The next logical step is a shelf that could handle 14-20 of these cards with plenty of fiber channel ports to hook up at least several servers. Will an 8GB fabric be fast enough? Until something like this happens, tiered storage is probably the way to go. I am sure they know this and are working on it. After all, The Woz joined the company.

Conclusion

I wish I could buy stock in this company.

email it! |   |   | 

Permalink     7 Comments

Geek Pr0n

email it! |   |   | 

Permalink     7 Comments

Notebook opinions?

email it! |   |   | 

Permalink     6 Comments

Drool: 4,200,000 IOPS

Found this while reading the SSD article on intel's new laptop SSD. The game is going to change and it is going to change fast. Hell, I might go buy some intel stock!

email it! |   |   | 

Permalink     1 Comments

Vacation Post: EMC Clariion CX4 support SSD Drives

A beach tip from a cousin in law that works at EMC. Here is the press release. I am not going to go into a lot of txt but this is the tipping point. Get ready for a wild ride. Soon 1GB network connections won't be enough! Related posts: SSD

email it! |   |   | 

Permalink     1 Comments

"Disks are dead, they just don’t know it yet"

I have been drooling over following Texas Memory Systems for a couples years and more recently, BitMicro.

I am not comparing the products that the companies sell because they are different products and the measurements are not the same. These bullets are mainly for drool factor and background for a post that I will refer you too.

Texas Memory Systems:

RamSan-400 SSD SAN

• The World's Fastest Storage®

• First solid state disk with 4Gb Fibre Channel interfaces.

• First solid state disk with 4x InfiniBand interfaces.

• Over 400,000 random I/Os per second.

• 3000 MB/s random sustained external throughput.

• Full array of hardware redundancy to ensure availability.

BitMicro:

E-Disk® Altima™ 4Gb Fibre Channel 3.5" Solid State Drive

• Up to 640GB of storage per disk on 1" drives.
• 1.6TB on 3.5" drives
• 800 MB/sec Full Duplex Burst Rate
• Up to 55,000 IOPS I/O Rate
• Similar offering on u320 SCSI

Now these are numbers from the manufacturers. Lots of missing info like read\write numbers. Numbers for different sizes of reads and writes etc. However, did I say WOW?

I am making this post because I ran across a blog post by Mike Ault on a FriendFeed conversation.

Mike address's these SSD "lies"

1. Solid state drive technology is very expensive
2. Solid state devices are best when directly attached to the internal bus architecture
3. Solid state drives will only be niche players
4. You can get the same IO rate from disks as from SSD

and he ends with this quote:

I am not afraid to say it: SSD technology is here, it is ready for prime time and it is only a matter of time before disks are relegated to second tier storage. Disks are dead, they just don’t know it yet.

We can only hope so :) I highly recommend reading the whole posts here.

I hope Microsoft is watching this technology. Sure, SQL will like a SSD SAN right now but I bet it could be heavily be optimized to run on SSD.

edit: link fixed

email it! |   |   | 

Permalink     3 Comments

Return of the 8 way

It seems like just yesterday Compaq Proliant 8500’s were the bomb with 8 – 550 mhz processors. 4.4 ghz of processing madness! That was even before hyperthreading and multi-cores. Since then, the big 3 server hardware vendors eliminated 8 way machines in their commodity server lines. You still had 8 way options but there were more cost effective configurations because multi-core processors removed the need for 8 way boxes in most cases.

That is until now as HP releases the dl785 g5. Eight sockets capable of running quad-core 2.3GHz AMD Opterons. That is a combined speed of 73.6GHz. They comes with 8GB of RAM but they support 256GB of RAM(512GB when 8GB dimm's become available). The servers, themselves, are going to be relatively cheap compared to the high end SAN's and large amount of memory needed to get the throughput high enough tax the processor sub system. Without a large spindle count\cache and amount of RAM, the system will have an IO bottleneck long before the processors in most cases. Of course, some applications have special needs. :)  

So when will these come into play? I think the biggest use of this box will be for consolidation particularly on SQL Server 2008. Imagine taking 20 or 40 instances on different OS’s, hardware, storage etc and making it one(or even 2 or 3) SQL Server 2008 instance. The environment would be so much easier to manage. The SQL Server 2008 resource governor was made for consolidation. Some of the new features in SQL Server 2008 are going to be CPU hungry like the spatial data, partition parallelism improvements and transparent data encryption.  The data and backup compression features push both ways by lowering IO and increasing CPU with the idea of decreasing execution time. Even if you go to SQL 2005, it would be a nice upgrade for a consolidation box.

Of course, you have to worry about having all of your eggs in one basket but that is another post.

Introducing the dl785 g5:

email it! |   |   | 

Permalink     2 Comments