Slashdot Mirror

.NET runs on puny Intel-like servers

.NET runs on puny Intel-like servers

C-JDBC can handle more than just full partitioning or replication, it also provides partial replication (a little bit like you would use RAID-5 with disks).
The idea is that with full replication you have to broadcast the write to all databases (to be consistent) and you can only balance the reads. By controlling the replication of each database table, you can have scalable performance. Look also at the nested RAIDb levels, it's pretty cool to build large configurations.
Some tests have been done with TPC-W and performance scales linearly up to 6 nodes (we did not have a larger cluster to test bigger configurations).
Sure it will not replace very large Oracle configurations at the end of the year, but it looks very promising.

By the way, if anyone is interested in TPC-C benchmarks of the Opteron, it can be seen here. It is interesting to see such a small player like Rackserver among big boys like IBM, HP and Fujitsu.

Those TPC-C Price Performance numbers are awesome but has anybody noticed that the rig runs $227,000???????

Damn... talk about Enterprise pricing!!!!!

Those TPC-C Price Performance numbers are awesome but has anybody noticed that the rig runs $227,000???????

Damn... talk about Enterprise pricing!!!!!

Sun is in big trouble. When you can buy a cheap fault-tolerant Opteron server from IBM, why would you spend twice the amount of money on a Sun server.

This statement is interesting to me, since I keep a eye on TCPC. I looked at the most recent results for non-clustered performance. I could be wrong, but I compared the results of the top three. #1 for non-clustered is a Fujitsu setup solaris 8 server from 2001. The Fujitsu setup used BEA tuxedo to schedule and handle the transactions. The microsoft setup by NEC used the clients to schedule the transactions. That is a fairly large difference and doesn't provide the same level of reliability or recovery. If a client dies, all the transaction on that server are gone and that doesn't provide state replication of the transaction across multiple clients. Scheduling the transaction on the client makes sense to reduce the load on the server, but it's a totally different architecture and isn't a one-to-one comparison. To effectively provide 99% reliability and recovery with the NEC setup is will be significantly harder because the clients now have to provide full clustering and state replication.

Although it's great that SQL Server has improved, the response time of the NEC system is 3x slower than #3 IBM. Another important fact is the NEC server was loaded up with 512Gb of RAM. The fujitsu system from 2001 only has 256Gb of RAM. It's no wonder the NEC system did so well. They simply loaded the entire database into memory. How much does that system cost? According to the summary 5,619,528.00.

If you're spending that much money, why would you even consider a non-fault tolerant architecture?

This statement is interesting to me, since I keep a eye on TCPC. I looked at the most recent results for non-clustered performance. I could be wrong, but I compared the results of the top three. #1 for non-clustered is a Fujitsu setup solaris 8 server from 2001. The Fujitsu setup used BEA tuxedo to schedule and handle the transactions. The microsoft setup by NEC used the clients to schedule the transactions. That is a fairly large difference and doesn't provide the same level of reliability or recovery. If a client dies, all the transaction on that server are gone and that doesn't provide state replication of the transaction across multiple clients. Scheduling the transaction on the client makes sense to reduce the load on the server, but it's a totally different architecture and isn't a one-to-one comparison. To effectively provide 99% reliability and recovery with the NEC setup is will be significantly harder because the clients now have to provide full clustering and state replication.

Although it's great that SQL Server has improved, the response time of the NEC system is 3x slower than #3 IBM. Another important fact is the NEC server was loaded up with 512Gb of RAM. The fujitsu system from 2001 only has 256Gb of RAM. It's no wonder the NEC system did so well. They simply loaded the entire database into memory. How much does that system cost? According to the summary 5,619,528.00.

If you're spending that much money, why would you even consider a non-fault tolerant architecture?

This statement is interesting to me, since I keep a eye on TCPC. I looked at the most recent results for non-clustered performance. I could be wrong, but I compared the results of the top three. #1 for non-clustered is a Fujitsu setup solaris 8 server from 2001. The Fujitsu setup used BEA tuxedo to schedule and handle the transactions. The microsoft setup by NEC used the clients to schedule the transactions. That is a fairly large difference and doesn't provide the same level of reliability or recovery. If a client dies, all the transaction on that server are gone and that doesn't provide state replication of the transaction across multiple clients. Scheduling the transaction on the client makes sense to reduce the load on the server, but it's a totally different architecture and isn't a one-to-one comparison. To effectively provide 99% reliability and recovery with the NEC setup is will be significantly harder because the clients now have to provide full clustering and state replication.

Although it's great that SQL Server has improved, the response time of the NEC system is 3x slower than #3 IBM. Another important fact is the NEC server was loaded up with 512Gb of RAM. The fujitsu system from 2001 only has 256Gb of RAM. It's no wonder the NEC system did so well. They simply loaded the entire database into memory. How much does that system cost? According to the summary 5,619,528.00.

If you're spending that much money, why would you even consider a non-fault tolerant architecture?

Ultrasparc places in the low, mid, and high-end server ranges on TPC-H, but IBM and HP will sell you a PC to compete with Sun on the low end, and IBM doesn't even place in the 3TB high-end systems. HP will sell you a PA-RISC system to compete with SunFire 15K. With IBM and HP, you have to rewrite software from the ground up to scale from low to high end of their Sun-competitive server offerings. It's not the same. Apples and oranges.

Oh, BTW: TPC-H results

At the low end, Sun finds a hard place. The story is pretty much the same. Sun's SPARC has extremely poor performance, compared to the x86 or its clones. So, Sun has contracted with Chinese manufacturers to build servers and desktops that run Linux atop an x86 or its clones. The trouble is that the profit margins are thin, and Sun faces severe competition from IBM, HP, and Dell. Sun does not have a significant services organization that can reap profits by providing contract software and support for these cheap x86 boxes. IBM and HP do, however, have significant services organizations that can generate profits from servicing cheap x86 boxes sold at a loss.

Now, given Sun's unethical business practice of preferring H-1B employees over American citizens, do you think that any self-respecting American business would consider Sun?

The Sun is setting. Good riddance.

At the low end, Sun finds a hard place. The story is pretty much the same. Sun's SPARC has extremely poor performance, compared to the x86 or its clones. So, Sun has contracted with Chinese manufacturers to build servers and desktops that run Linux atop an x86 or its clones. The trouble is that the profit margins are thin, and Sun faces severe competition from IBM, HP, and Dell. Sun does not have a significant services organization that can reap profits by providing contract software and support for these cheap x86 boxes. IBM and HP do, however, have significant services organizations that can generate profits from servicing cheap x86 boxes sold at a loss.

Now, given Sun's unethical business practice of preferring H-1B employees over American citizens, do you think that any self-respecting American business would consider Sun?

The Sun is setting. Good riddance.

Actually, Sun's SPARC processors have performed horribly on the TPC-C benchmark by the Transaction Processing Council (TPC) . In fact, the performance of TPC-C on Sun's computer systems is so bad that Sun actually withdrew participation from the TPC-C benchmark. Today, that benchmark is dominated by outstanding computer systems built by the likes of IBM, HP, and (incredibly) Fujitsu.

Check out #2

Microsoft just set the #2 TPC-C result in the non-clustered category using Windows Server 2003 and a 32-way Itanium 2 machine. They did this, of course, because Oracle publicly derides clustered results as not counting (and really setting up horizontally partitioned views across a huge federation of serves is not the easiest thing, and it's far from transparent for the database developer: You have to specifically design around it), so now there's a SQL Server 2000 result higher than any Oracle result.

So there you have it: A 32-way machine that's actually useful (when available on 2003-06-30).

Ellison's prophetic comments, much like Scott McNealy of Sun, are generally worthless: If one looked at his historical claims they would find an astoundingly poor accuracy of their predictions. At some point shouldn't someone call him on his abilities as a seer?

The most ridiculous part of his comments that immediately pointed out how uninformed and idiotic his vitriolic claims are is the statement "They had a virtual monopoly on Web servers, and then they were wiped off the face of the earth. And it's going to happen to them again on Linux.". The Slashdot summary itself points to the Netcraft graph, but strangely fails to points out the absurdity of Ellison's statement: Microsoft has never had a "virtual monopoly" on web servers. Indeed, Microsoft was an underdog, came into the game after Apache, and has grown to 28%, gaining 5% or so during a period when Apache marketshare has remained constant.

P.S. Ellison is going to have to develop a new angle to push Oracle - When SQL Server trounced them in the clustered results on the TPC-C, Ellison and friends proclaimed that clustered results don't count, getting the TPC to allow one to separate clustered and non-clustered. Well now Microsoft beats Oracle at non-clustered results too. I'm sure there'll be some new angle to defend against this.

What a joke.

Also, windows (2003, running on a 32-processor machine) holds 2nd position in the non-clustered results. And while the performance is within 95% of the leader, it's price is less than half.

What a joke.

Also, windows (2003, running on a 32-processor machine) holds 2nd position in the non-clustered results. And while the performance is within 95% of the leader, it's price is less than half.

Not to answer for him but, here's one for file and printer sharing. There are also a number of TPC benchmarks that show Linux outperforms MS W2K running Oracle on identical platforms.

That said, I'd like to second your point that broad statements like "outperforms" should always be in the context of "at what". It's like me saying "I'm faster than George" ... it just doesn't mean anything when it's out of context like that.

Not to answer for him but, here's one for file and printer sharing. There are also a number of TPC benchmarks that show Linux outperforms MS W2K running Oracle on identical platforms.

That said, I'd like to second your point that broad statements like "outperforms" should always be in the context of "at what". It's like me saying "I'm faster than George" ... it just doesn't mean anything when it's out of context like that.

If IBM found compelling reasons to do so, then said compelling reasons must exist. Now, you don't see AS/400s at the top of the charts at tpc.org, so I would tend to believe that it isn't necessary for high performance, but that doesn't mean that there is no benefit at all.

p.s. With no views, no triggers, and no foreign keys, I wonder why MySQL gets all of this hype. There are so many free databases that implement these features and don't seem to suffer from performance issues.

BEA is already competing against an Open Source product-and loosing the battle. Oracle could be a fairly early casualty. A big chunk of the appeal of Oracle rests on its ability to compete in industry standard benchmarks. Open Source products have historically tended lang behind in having "glitzy" interfaces-but have tended to excel in reliabilitly and performance. It is clearly a logical development that in the next few years, the Oracle database will find itself replaced by an Open Source Database-this will be a tremendous blow to the prestige of Oracle as a company.

Now, most Oracle revenue comes from sevices and various accounting programs-but there are also starting to emerge various Open Source Accounting packages(i.e. SQL Ledger) that might in time start to hit Oracle more directly in the pocketbook.

Besides the TPC benchmark for clustered machines, already mentioned in other posts, there is also a result for a non-clustered machine, HP rx5670 (4 Itaniums 2.1GHz) with Oracle 10i. Same performance than hp/ux on the same platform, at a fraction of the cost. See both results here and here

Besides the TPC benchmark for clustered machines, already mentioned in other posts, there is also a result for a non-clustered machine, HP rx5670 (4 Itaniums 2.1GHz) with Oracle 10i. Same performance than hp/ux on the same platform, at a fraction of the cost. See both results here and here

Acutally, it does

Right here......

in the top 5 for clustered performance.

...it is there, 4th position in the price/performance clustered solutions:

www.tcp.org/...

RedHat AS, running Oracle 9i

The framework is a good start, but Linux would not be considered enterprise-ready until it actually appears in TPC-C results list (at least in first hundred).

For those that didnt understand what TPC is and did not get enough information from http://www.tpc.org, check out further information that is very well explained at The Benchmark Handbook. The direct link is at http://www.benchmarkresources.com/handbook/content s.asp.

Suhit

But, isn't one of those situations he mentions in the interview (namely, running a large database server) what this chip is designed to be doing?

Sure, but it doesn't really do it significantly better than some of the more common RISC architectures (Sparc, Power, Alpha), and it's a lot more expensive.

Do you have any data to back that up?

According to this press release, the current record for the TPC-Cbenchmark on a 32-way machine is held by an NEC server with Itanium 2's.

Take a look at the latest tpc-c scores. The top ten non-clustered servers in terms of performance has always been one of the most coveted positions to have your server in. Notice that Sun doesn't even make the list. Not even their super-fantabulous E15K!

Top Ten Non-Clustered TPC-C by Performance Version 5 Results

Of course Sun internally runs these benchmarks, guess why they don't publish the results. BTW, I've heard that optimistically the 72-way E15K scores about 315,000 tpmCs...pathetic in comparison to servers like the 32 way IBM p690 or the 32 way NEC 5800 which do way more for less.

Now slowly read between the lines of my message and you'll see why Sun is already starting to extinguish.

I'll acknowledge that this type of benchmark is a perpetual horserace, and the leader will always be changing. But the Sun horse threw a shoe and broke its leg about 2 years ago! It's gonna get shot within a few years...

"If microsoft can't get windows ... up to enterprise class"

Linux's clustering capabilities are indeed better than those of Windows, but only in the engineering and scientific calculation space.

You seem to be overlooking the enterprise database space, where Microsoft has thoroughly smacked-down the competition, both in overall performance and price-performance.

For "enterprise" computing, what is more important: scientific calculations or databases? I think you will find the latter more critical to the overwhelming majority. Many, if not most, enterprises do not perform the kinds of engineering and scientific calculations that grid computing targets, while most would be hard-pressed to find a company that does not use a database.

I'm not trying to ridicule the apparent success of linux in this space, but don't delude yourself into thinking that this is the be-all and end-all of computing just yet.

Hmmm .... tpc.org is an interesting organization. It is a non-profit who is funded by memberships from the hardware/software companies on which it produces benchmarks.

According to their website, "Full Members of the TPC participate in all aspects of the TPC's work, including development of benchmark standards and setting strategic direction. Full Membership costs $15,000 per calendar year."

Wow, a large percentage of the benchmarks are using MS operating systems. Oh look full members get to set benchmark standards. Mmmm, the only pure OS company who is a full member is Microsoft. I wonder what kind of conclusion can be drawn .......

Go to the TPC website and take a look at score reports for the TPC-C benchmark, which is an online transaction processing (OLTP) benchmark going back 10 years or so.

Score reports for most mid-end IA-32 quad-processor servers reveal that they are using several four-channel Ultra-160 SCSI RAID controllers, and fifteen hard drives per channel. My professional experience with TPC-C shows that the hard drives' throughput get maxed out way before the SCSI channel bandwidth does, and we're talking 15 drives per SCSI channel. That's why these benchmark results are still obtained with Ultra-160 controllers and drives instead of Ultra-320. The extra bus bandwidth of Ultra-320 SCSI doesn't buy you much because the fastest disks out there cannot churn out data fast enough to max out a Ultra-160 interface.

I was recently looking at both IDE and SCSI drive specs on manufacturers' websites. I saw Ultra-160 and Ultra-320 SCSI devices with seek times of 3.5 ms and rotational speeds of 15,000 RPM. But most IDE drive families are still at 7,200 RPM max and have seek times of 8.5 ms or more. The better seek times and rotational speeds are the main reason I would upgrade my storage to SCSI (if the costs were not so high, that is :-). The product reviewed here provides exactly the reverse of the functionality I want. As such, I think it's useful only for specialized applications like putting an IDE CD-RW in a SCSI-only workstation or server.

The TPC-C benchmark is very good, and there are enough multi-cpu results that this should inform you a bit.

In general, for generic database work the important system aspects are:
-very large I cache
-lots of main memory bandwidth
-lots of secondary store bandwidth

This would suggest that a 2x xeon would be better than a cheaper 4x P3 box. While it's impossible for us to give you sizing advice without a more detailed question, my guess would be you need 2 xeon boxes with 4 SCSI disks each in RAID 0-1. One should be enough to run your production work, and I wouldn't personally run MS SQL2k without having a hot backup running as well. It's reliability is a bit better than say, MySQL, but it still will complain now and again. With intel kit, the things that will fail most often will likely be disk controllers and the like, not the SCSI disks, so I'd even chose having a hot backup system over a more reliable RAID setup.

Here is a link to the Defence Finance and Accounting Service http://www.dfas.mil (Yeap .mil)
http://www.dfas.mil/technology/pal/dcii/dcii-proce dures/dcii-doc/ept.pdf Otherwise a Complete Testing/Performance Procedures for Many Different Approachs are described at the http://www.tpc.org
Test TPC-C TPC-H TPC-R and TPC-W
You will find a description of all of them here http://www.tpc.org/information/benchmarks.asp and each have their own pdf description. VERY FORMAL and detailed. Have Fun ])

Here is a link to the Defence Finance and Accounting Service http://www.dfas.mil (Yeap .mil)
http://www.dfas.mil/technology/pal/dcii/dcii-proce dures/dcii-doc/ept.pdf Otherwise a Complete Testing/Performance Procedures for Many Different Approachs are described at the http://www.tpc.org
Test TPC-C TPC-H TPC-R and TPC-W
You will find a description of all of them here http://www.tpc.org/information/benchmarks.asp and each have their own pdf description. VERY FORMAL and detailed. Have Fun ])

> ... a Power4 processor MODULE consists of eight processors

True, but it still counts as one CPU for the SPEC benchmarks. (Note: the 32-way Power4 is a BEAST on TPC-C ... 5% slower than a 64-way HP superdome and only 13% slower than a 128-way Fujitsu Primepower. )

By the way, your article goes on to show that the Power4 is about 65% faster (per clock cycle) than a Pentium 4 for SPECint2000, and it's roughly tied (+/- 10%) with Itanium and Alpha.

Yeah, and so far SQL Server is the most scalable database server on the planet.

http://www.tpc.org/tpcc/results/tpcc_perf_results. asp

M$ has absolutely NO business badmouthing ANY other vendor's database product. M$ SQL is the biggest POS I have ever seen. DB2, Oracle and Sybase beat its pants off all day long.


Oh, really? Then why does the TPC say that MS-SQL + IIS holds the top 9 spots for e-commerce performance and the top 3 for transaction processing?

M$ has absolutely NO business badmouthing ANY other vendor's database product. M$ SQL is the biggest POS I have ever seen. DB2, Oracle and Sybase beat its pants off all day long.


Oh, really? Then why does the TPC say that MS-SQL + IIS holds the top 9 spots for e-commerce performance and the top 3 for transaction processing?

1. MS IS scalable. Don't believe Microsoft, check out TPC. Microsoft virtually "owns" several benchmark categories, especially in clustered configurations.

2. Most of SQL's security problems had to do with default sa password. The same problem exists with Oracle, IBM, etc. It's a matter of the DBA fixing those holes "first thing".

For most everything else with really big requirements, Windows just doesn't make sense, whether for reliability, scalability or performance.

Sounds like someone's been fed some tall tales from the old grizzled Unix gurus. For the truth on transaction handling, check out the Transaction Processing Perfomance Council's figures. Oh my goodness, Microsoft's SQL Server rocks them all! How can that be? When you throw as much hardware at it as the mainframe guys do, it blows them all away, because its per-processor performance is higher than all the rest. Oracle has 6615 tpmC per processor, while SQL Server has 9644 tpmC per processor. Tell me again about Wintel's lousy transaction capabilities?

Is there some reason that the article mentions that the Linux solution "beats" the Windows solutions, but everywhere on the site the Windows servers (whether it be IBM's DB2 or MS-SQL) are the leaders of the pack many times over?

That article link is like something a politician would say. Do you realize that Oracle is optimized for UNIX? It never should have been ported to Windows. Such a comparision is akin to saying "Apache 1.x runs better on Linux than on Win32!" No shit, Sherlock. Anyone who looks at the top ten results will clearly see that the point this article is trying to make means jack as far as the real numbers go.

By the way, I hear the latest Linux kernel is better at running Linux programs than CygWin!

Is there some reason that the article mentions that the Linux solution "beats" the Windows solutions, but everywhere on the site the Windows servers (whether it be IBM's DB2 or MS-SQL) are the leaders of the pack many times over?

That article link is like something a politician would say. Do you realize that Oracle is optimized for UNIX? It never should have been ported to Windows. Such a comparision is akin to saying "Apache 1.x runs better on Linux than on Win32!" No shit, Sherlock. Anyone who looks at the top ten results will clearly see that the point this article is trying to make means jack as far as the real numbers go.

By the way, I hear the latest Linux kernel is better at running Linux programs than CygWin!

Is there some reason that the article mentions that the Linux solution "beats" the Windows solutions, but everywhere on the site the Windows servers (whether it be IBM's DB2 or MS-SQL) are the leaders of the pack many times over?

That article link is like something a politician would say. Do you realize that Oracle is optimized for UNIX? It never should have been ported to Windows. Such a comparision is akin to saying "Apache 1.x runs better on Linux than on Win32!" No shit, Sherlock. Anyone who looks at the top ten results will clearly see that the point this article is trying to make means jack as far as the real numbers go.

By the way, I hear the latest Linux kernel is better at running Linux programs than CygWin!

Of course, I'll get flamed for not bashing microsoft

Don't be silly. :)

From this FAQ you'll see:

In general, TPC benchmarks are system-wide benchmarks, encompassing almost all cost dimensions of an entire system environment the user might purchase, including terminals, communications equipment, software (transaction monitors and database software), computer system or host, backup storage, and three years maintenance cost. Therefore, if the total system cost is $859,100 and the throughput is 1562 tpmC, the price/performance is derived by taking the price of the entire system ($859,100) divided by the performance (1562 tpmC), which equals $550 per tpmC.

Most people would focus on the hardware cost, but in reality the highlighted maintenance cost took the precedence.

Most midrange UNIX server has outragous maintenance cost. The maintenance cost of a UNIX server in the third year could be exceeding the cost of the hardware itself. It's due to the fact that older parts are difficult to find, thus make maintaining older servers more difficult. Besides, they really want to cut older production lines in favor of newer servers production.

x86 platform is known to have flat and lower maintenance cost, due to the low cost hardware and high compability with older hardware, i.e. older parts can be found easily. That's why Microsoft could easily beat the TPC pissing races.

Doomsday finally comes to Microsoft when Linux is entering the database market. Although at this moment big corps are still offering Linux maintenance with cost comparable to UNIX package, that's not surprising when Linux engineers are not as abandon as MCSE. However, it'll not be the case in the future. I think Microsoft would eventaully lose this pissing race in the long run.

If you look at the top results in a price/performance comparison, Windows 2000/SQL server are still winning.
http://www.tpc.org/tpcc/results/tpcc_price_perf_re sults.asp?resulttype=cluster&version=5

It's interesting, though, that Red Hat had the cheapest Oracle implementation.

well, then it's only one thing to do; run the tpc test on each database and see whichone will perform best.

Windows 2000 with SQL Server not only tops the performance charts, but it also owns every single spot on the performance/price charts.

free!=better