Study Notes

Sol-box’s mail server contained such device called Flash-RAM,which has a mini Linux kernel.The Flash RAM is inserted in IDE slot,and always is smaller as being compared with harddisk.Once the kernel in the IDE-ROM startups, it will detect all the modules and initiate the drivers of devices.The data is stored in the harddisk,and the kernel tell the computer to operate with specific devices,such as harddisk.When harddisk halted,it will give the following message:kernal panic and then we should change another harddisk.So the kernel is the soul which is the most important.Kevin said there are about 70 engineers in RD(research and develop) team.They used the kernel of 2.4.30 to product a Linux system named Bax.The Bax includes drivers of ethernet card and Raid controllers.Almost these drivers are based in Intel platform,but do not contain every devices pubulished by Intel.For example,Marvell gigabyte ethernet card.

In many cases,the FAE would be asked to diagnose in console mode.Telnet command is commonly used.First,we should called the customer’s system administrator(In Taiwan,called MI meaning Manager Information),to open port 22 which is for SSH.Perhaps MI has never known about how to configure the firewall to do this or MI is worried about the whole system’s security,they would not open the port.How to check the port 22 has been opened? In the command line statue,issue the following commands:

#ping 192.168.1.26 22

If ssh messages display after pressing Enter,it means the port has been opened successfully,then engineers could connect to the server remotely by telnet.

When the mail is refused by the mail-server, there will be a responsed mail sent to the user.Just image,the reminding mail is sent as soon as the server find rubbish mail,how many remind mails will be in your mail-box?It even becomes another type rubbish mail.So we should configure the mail-server to hold these mails and sent them together to user in specific clock.

The error message of sending email must be in English language for convenience.In the email’s detail mode,text following by title “subject” should be coded for protecting user’s privacy.However,about 70% mail-servers are not in this way.The text in the email’s header contained more than 1500 words is also considered as rubbish letter because ofbehavior which is not common.DoS,and Dictionary attacking usually happen.We can take corresponding strategy to prevent them.For example,MI could filter the IP which tried to send mail by using dictionary if the connect counts are more than three times.

Behavior and culture.Although culture is very differnt, it’s not realistic by using either Black-up strategy or Content Filter.The only way,or the beat measure to prevent rubbish mail is using Behavior judgment.A mail is judged as rubbish or not does not rely on its contents, the behavior of sender Mail-server is our concern.

Kevin,senior engineer from TaiWan,gave us a lecture about mail server producted by his company in Shanghai.

Internet mail,also called e-mail, are invented by some teachers in US, they uses email to exchange message with others as fast way.Twenty years passed, email became more and more important communication tool not only in college but also in business activities.In 2002,experts considered the security of mail was most important,so they work hard in encrypting package flowing in the network.but with the time goes into 2003,something changed,the mail users found a large number of letters contained information they did not want to read had been sent to their own mail-box.It was very annoy.The rubbish mail saved in the harddisk of mail-server causes waste and limit the transimition of network.So people dedicate their life to computer had to resolve this trouble.In 2004,many products about preventing rubbish mail were developed.

There are total three strategies, behavior,black-out,and content filter.

Behavior strategy collects the senders’ information and analyze their behavior, then help us to decide whether receive or refuse.This is the most popular mode.Black-out,which is also called Black name list,filters those mail from the address in the black name list.Such list would be created and update by Mail server administrator or by uses themselves.Content filter,reading the whole text of mail,and compares each word with those in the sepecial dictionary,once matched,the mail-server judged these mails as rubbish and refuse to receive.In fact,it is not a good ider for preventing rubbish mail because perhaps this strategy may refuse the mail which is not rubbish.We always find that the employees of the supermark distribute the colorful ad.paper,Do they break the law?The answer is denial.So does the email.But we must do something firmly to protect people’s privacy and prevent wasting resouces.

How do users access the specific website? The packages are sent to many routers.Routers is such device that direct next step or a good man tell you where to go next.It sent from router to router until find the specific website.Note, what is the difference bwteen http and https?Https means security http transmit by using SSL(security socket layout).In contrast,http does not have high security, because the sniffer installed in some routers may catch the information of packages.

For culture is very different all around world,one same mail,some people may think it is useful,but some people may refuse. So mechanism of mail server should also contain some mail which are judges by users themselves.The uses can read it or delete it without be effected too much.

Kevin said there were many rivals,we should let customer know our pruducts also contain rival’s advantage and surely instead them,or the sales told the Project Manager to improve our products in the next generation.However,we must not give the customer any information about new product in developing.That’s crazy.

Queue is the standard of judging mail server’s performance.A good mail server system has enough queue and send mails as soon as possible.It’s powerful.If the queus is not suitable,it will cause many troubles,for example,the mail is always waiting in the queue, it will never be sent.

SMTP server:snipper(for backup),relay(the most popular,about 90% in this market).

DNS constructure:
-Authority:resolute name step by step liking router(eg,abc.com.cn,cn –> com –> abc)
-Resolution:have a large database to record resolution items,the database is updated as finding new resolution recorder,so it becomes more and more bigger and intelligent.

In the SE7520BD2 platform,when I inserted USB disk any issue the following command:

# fdisk -l

I did not find usb device.So,”lsmod”, and reload the USB modules,”rmmod –> insmod”,and then
execute this command as root:

# cd /mnt
# mkdir a
# mount /dev/sda1 /mnt/a

Then,USB device was installed successfully.

Source code means everything.Once you have the source code, you could modify and re-complie that code wherever you want to use,any situation, any time.
I was very tired last night,and went to sleep at about 22:00.Hmmm,it was a wondelful sleep and I feel I’ve had good rest to continute my work.
In the morning, I heard PR1580 had been cut. I was so surprised that asking Weining Qi,product engineer of PR1580.Qi said Project Manager had resigned one month ago, so this project developed with MSI was cutted with unknown reason.He was also sadly because both of us have put all our heart in this subject since Mar 19th.I think Qi is a good associate,working with him is in high working-rate and spirit.
How long do I not enjoy football? Maybe one year. This afternoon, I called our captain,Zhaoyong Mo.In the 4th village of Melin, we played football with some policeman.I found myself in the ground,enjoyed football game.I love blue sky,white cloud and green grasses.Mo gave me three nice through pass,and I soccered three goals.The most exciting moment is,when I was in left back-ground,I hit a long pass to our mate in the font-right area,he received the ball very easily,however he failed in making a goal.
I went to Carry Four supermark with my mother when I got home.I brought a book,”legend of Hongzhang Li”, writen by Qichao Liang, a famous historian.Li has been argumented by people for many years, some people consider he selt motherland and is a shame,but he is a wisdom man in another people’s eyes.I do not care whether he is good or not, I want to know his thinking in those worst years.Li is a legend,influenced China and world deeply

In the training class, Jiang said that using Intel’s MPI complier would improve 700% work rate than general complier(gcc). 411 is an secure tool for monitoring cluster service.

Two useful Linux commands, after having modified the fstab or mounted file-system, the following commands should be issued:

# service autofs restart

If we failed in “make” and wanted to recomplie(make) again, we should execute the following command first to delete former complied files which are not useful any more:

# make clean

ROCKs has its own database which records all the actions of cluster. The manual of ROCKs is “NFCA ROCKs cluster usr guide”. Now, the current release of ROCKs is 3.3.

All the above contents are included in the lessons of last date, I forget to write here yesterday. The following contents are todays’ lessons.

Parallel Program Design Methodology, including partition, communication, agglomeration(make all strips together) and mapping.Partition will cut the working load into many strips, just like grid. all these grid will communicate with each other, and they will be made together after computing complete.At last, assign tasks to processors in order to maximize processor utilization and minilizae interprocessor communication.

Distributed and shared memory are two type of parallel program strategy. Distributed nodes are independent and have own memory. Shared memory will share public memory. We can refer “Designing and building parallel program” writen by Lan Foster. Mr Jiang said this document perhaps would be downloaded for Internet.

Computations will be partitioned by Function or Domain.
– Functional decomposition: Divide the computation, then associate the data.Focusing on computations can reveal structure in a problem.
– Domain decomposition:Divide the data into pieces, then associate computation.Focus on largest or most frequently accessed data structure.

The most difficult thing is finding data parallelism.In some case, parallel is not suitable,for example, serial compute liking we could not do input and write parallel, if so, the result must be wrong. For database, data must be commited before be writen in the harddisk.

In the cluster’s computing, race always happens. Programmer use Debugging function frequently to find error in their code, however, during Debugging process, racing are all ignored.It is annoy in the paralle computing because racing really exist in the cluster.We’d better present Itanium platform to customer with Intel’s complier which can maximize the Itanium processor’s function. Tips: parameter “spec” indicate the performance of processor(CPU).

Gaussion, used for chemical computing, forbidden its source code in China area, so we can not improving performancing by Intel’s complier,thourgh optimizing hardware structure would get litter advantage.

The most common MPI implementations is MPICH. Some customers ofen complained that the hardware platform ran slowly when the program was writen by customes themselves or un-certificated in a few years. In fact, it was the non optimize program code causing this matter.We should tell the customers earlier with great patience.

In the Cluster which run computing, if there is one node haltes, then the whold computing failed. We could only check the halted nodes and restarts computing again when ensure every node is good. Is the cluster not stable? For example, there is a work which need one week computing time in one machine, now, imagine making the work computed in a cluster contains 4 nodes, the time will decrease sharply to 7 or 8 hours.If one of the four nodes failed when computing, we have to restart this job with more 7 hours, not waiting for a week. So, the example indicates cluster improves work rate greatly though it has a litter matter.

A Linux command line: find -name ‘file_name’. its purpose is to find out specific file usring file_name string. I think this command does not do better than “locate” which is more convenient for type.

HPC Benchmarking. The sales are always asked for Benchmarking of the product by the customers in whose mind Benchmarking indicates the performance would help them to select from so many products. In the Lab, we will tuning the system’s structure, for example, If we have four processors(CPU), there are three strategy of distribution. One schema is 4 processors are all on one node, if each node has one processor is also included in the strategy, and perhaps for the best performance, we put two cpus in two machines so that each node has two processors. finally we can determine which hardware to purchase.

What can be done to improve the Benchmark?
—— Modify the data.
—— Tune the application, such as better complier, better libraries, and modify the benchmark code itself.(There are many libraries in the market now, netlib.org and GOTO are opensource, but MPC from Intel must be purchased for business license.However, the former two lib do not supply unbreakable support as Intel’s MPC)
—— Change the hardware.(As we all know, the best rate of EtherNet will be 50-60%,and the InfiniBand will get about 70%). If more money is costed on nodes, less money is costed on communicating hardware, for example, switch, copper line of EtherNet. On the contract, more money is costed on hardware, less money is costed on nodes).

In the Intel trainning camp. HPLinpack is used to indicate Benchmark. Before HPLinpack test start, the Hyper-threading must be off because Hyper-threading technology confused the HPLinpack software.

To tuning performance, the litter value could be ignored or not is usually important to improve performance. we should test and test to find the right answer.

Cluster management tools packages include Portable Batch System(PBS),LSF and SGE. PBS and SGE is open-souce. LSF comes from Canada. In the traing lessons, we put attention on SGE, Sun Grid Engine. which originally developed by Gridware, a company departed from SUN.
We could see the masters and nodes’ statues in the Web brower(Moziila). There is not direct iron in ROCK linux, we run the moziila through click “run applications” and input the localhost IP address. Then three nodes’ details are all displayed in the Web Page writen by PHP.
In the items listed in the former web page, Ganglia is contained because it is installed by ROCKs automatically. Ganglia is used to monitor the cluster in GUI environment.

Before I came here, I was confused about this subject.
Having listened to your lecture, I am still confused, but on a higher level.

——Enrico Fermi

What’s HPC(High Performance Computing)? HPC is based on concurrency that computing in which multiple tasks are active at the same time, and will be used in solving bigger problems or fixed-size problem in less time. For example, scientific computing in physical, biologic or chamical subjects which both have may data need to be access and computed, we could not waiting very long for the computers completed this work. It seems that HPC would solve this problem perfectly.Althought adding more cpus would speed up, but the whole cost become very expensive. The nodes of HPC was common so that all the hardware of the nodes could be brought in the market.In this way, we get better performance with less money.What’s cluster? Collection of connectors, independent computers which work together to solve a problem. When we talk about clusters, the stack including environment (such as location, power,cooling ,sercurity )and assistance (Trainning) must be considered.There are many cluster solution stacks for Linux. OSCAR, NPACI-ROCKS and Score are open-source software, and Scyld and Callident must be purchased for license.

Cluster hardware desire:Interconnect, 100Mps Ethernet, 1Gbit/s Ethernet, SCI, Myricon Myrinet,InfiniBand,Quadrics QsNet(Pay attention to InfiniBand who has another name “IB”). Processors(CPU), all Inter serie processor including P4, Xeon ,Itanium would be used in the Cluster, however, Intel recommends Xeon and Itanium(Tiger 2 Platform). Pentium processor only support 1-way platform, they will cause net constructure more complex.

For example, painting a picket fence, more people can finish painting soon, but all the processing time also including prepairing(loading data) and cleaning up(consume the data) time, we should buy the painter, and clear workspace when the work completed. All of this need time which could not be ignored!So the speedup has a limit pointe when the load and consume time are zero.It is not realizable.
indicate that we could not improve speed only be adding painers. The right way we should found is deciding good price/performaning value.

Load Balancing, when the opera was going to end, some fat woman came out to sing. The load balancing are determined by the slowest nodes because only after this nodes finished working the whole cluster is considered having completed working.Load Balanceing Granularity, when Application granularity matches hardware granularity the cluster gets the best price/performance situation.

Some useful Linux commands:

#server network restart   (when edit the network configuration)
#vi /etc/sysconfig/network-script/ifeth0      (configure the eth0) 
#halt -p  (shutdown the system with power off)
(mount DVD ROM, Linux recognise the DVD as SCSI device)
#mknod /dev/sr0 b 11 0  
#mount  /dev/sr0 /mnt/cdrom

How to install Rocks?
First, insert Rock based CD, and press “f” when displays Rock’s installation logo. In the command line, input “frontend”, then, the Rock will be in setup statues. It will ask for ROLLS which is package of packages in the fact. We input the following three CDs in this sequence “”BASE,HPC,Kernel,SGE”. the installation program runs automaticlly after loading the ROLLS needed. This is called master node.In the second machine named node1, we insert the BASE cd in the DVD ROM, and it will install automaticlly without any user’s action. When installing, the statue list in the master node will display node1.If we have no BASE CD, how could we do to install another node named node2? OK, power on the node2, and press F12 when bios posts. It will come in the PXE starup instance. As same as node2, the installation program runs automaticlly. We need to do nothing so we should drink a cup of water or have a short rest.It is convenience for building the nodes in fast way.

In the morning, Mr King gave us training about memory including DDR and DDR2.

The differences between DDR and DDR2 is DDR2 run faster and has lower working voltage. Some type memory named Rambus, assumed by Intel in 2001, encapsulated in another way.Though Rambus failed in Chinese market in the past three years, many experts thought it will be useful in the future computer systems.

Samsung, Hyrix are the first line manufactor which have good reputation, their products sells very well. In fact, all memery are product in the same working line, however, just like the CPU, they are tested and given a mark as 400Mhz if it is very good, or 333Mhz if it is just OK, and so on.

Registy Memory has register chipset on the PCB which maybe 6 layers or 4 layers. The register chipset buffers the data put in the memory so that Memory will work more stable if a large number of memory would have been requested.

Linux is the best suitable operating system for oracle, only the following OS was certificated by Oracle:

Red Hat Enterprise Linux 4 (RHEL 4), RHEL 3, RHEL 2.1,

Novell SUSE Linux Enterprise Server 9(SLES 9), SLES 8

if recomplie the kernel or use the third-party tools to modify the kernel, Oracle does not supply the unbreakable support. Oracle said there are so many Linux distribution all around the world, it is impossible to certificated every Linux distributions.

As we all know, Tencent Co. used Slackware Linux, which type of database would be instaledl in their servers? DB2, or Sysbase?

I’ve never seen IBM server with AIX through I heard many people said IBM machine was huge and stable.

Load Balancing test, there are totally seven servers as node. 192.168.1.231 acted as the load balancing server,and the others are really workstations whose ip is from 192.168.1.232 to 192.168.1.237.

238 was the loading server which supply the testing pressure.

First, install MS SQL server 2000 in every realy workstations, and let the sql server startup when completed installation.

The engineer from Nanning decided the strategy of cluster and send commands to our server(231) thought moderm which called remote control. About 40 squel instants was started to connect virtual database (192.168.1.231).

In the 231 machine, I could see the statue about connecing and re-connecting,192.168.1.231 is always the server node,and it distributed the load to the other node such as 192.168.1.232 and/or 192.168.1.233.

There are many strategy in the software which is sent to test in the lab. One strategy is that the PC or customer workstation will always connect to the specific node once it was firstly distributed to the node by the 192.168.1.231. Because some people may do some working in one of the nodes, when the node halted, the data the user are working were not committed, so it perhapes lost. we could not imagine what would happen if the important data lost. this accident need the strategy that ensure custmer always connect to the specific server.

Another strategy is random. The customer’s connect will be distributed to random node. With the count of connecting adds, the loading will be balancing automaticly.

By the way, how to setting the display parameter to 1024×768 in Linux? Only changing the parameter in the display dialog is not enough. Login as root, in the /etc/X11 direction, edit the X1186config file by using vi, find the line including”800×600″ “640 x480”, and insert “1024×768” in this line. saved the file, and logout.

Everything will be ok when login as root again.