2007年11月 的存档

Riser Card and Network bandwidth of SS2280

2007年11月29日 没有评论

In the morning, I had a meeting with Shan Xiaoyu(new comer in product department) and a testing engineer, Hou Xianghua. First off, I reviewed the BX1200 development process and pointed out something should be carefully concerted about. UIT OS installed in BX1200 is 2.6 kernel while the third party video-monitor software is implemented in 2.4 kernel before(RHEL 3.0 and Red Hat 9.0), so it seems to have to re-compile the codes in our 2.6 kernel to ensure it can run smoothly based on BX1200. Gavin told me, he wants the BX1200 becomes a general platform, either of end users or the software company who will integrate their monitor program and get benefit by reselling BX1200. UIT OS has not X-Windows, so it is not possible to view graphics in BX1200 console. The last point we must confirm is the Web management port. The current Web port(NAS/iSCSI) is 80, which should not be duplicated with Video-Monitoring programs.

SS2280 always annoys me. I found the platforms shipped in November 8th, have not riser card in the chassis. However, in the earlier box we got in October, it is implemented with two riser cards(full-length and half-length). As the result, we can not insert Intel 1000Mb Ethernet card in SS2280. I’ve reported this situation to Tobey Chan, QE from Inventec.

Tan yinhong said during his performance testing on BX1200, one network ports can get the best performance nearly about 1Gbps, while two ports still retain the same level. Obviously the result is not reasonable. The latest news I got from him, maybe it is caused by the re-direct codes. I call Inventec technical engineer, York Chen. The kindly man told me the two ports in SS2280 both have dedicated bus connected with ESB2 south bridge. Hmm, It sounds I have to validate the actual network bandwidth in tomorrow.

分类: 科技 标签:

Single nfsd thread/Inventec SS2280 issue

2007年11月29日 没有评论

In order to resolve UT issue, mounting big volumes is very slowly and the customer set the maximum time no more than 3 seconds. In fact, we find that every mount version on 2.6 kernel will mount the volumes very quickly. However, if the kernel is 2.4, it is expected to have pain. We do not want to modify our code in kernel level, so that have to test the whole system performance when nfsd is running, though usually 4-8 nfsd threads are default. The only reason is single nfsd thread could not cause “dlm_sendd” eat all the CPU resources.

Five servers which implemented with Gigabyte network device, and all the cables are connected to a dedicated Cisco switch in order to avoid network traffic. Every server held 10-12 iozone threads to write data across two NAS 880 nodes. The performance is quite good,single node “bo” is around 25MBps, and during the 3 hours continuous iozone pressure testing the node CPUs usually have high free idles.

It seems the single nfsd would resolve the UT issue now. However, I need confirmation and agreement from Gavin Huang. Thanks Lin Peng and Zhu Feng, you have done great job all these days.

Inventec issue: Their engineers will arrive Shenzhen tomorrow morning to fix SS2280 platforms.

分类: 科技 标签:


2007年11月27日 没有评论

OEM: Original Equipment Manufactor
FRU: Field Replacable Unit
BMC: Baseboard Management Controller
EEPROM: Electronically Erasable Programable Read Only Memory
NMI: Non Maskable Interrupt

分类: 科技 标签:

Linux Patent/Intel 45nm High-K Tech

2007年11月25日 没有评论

It is not Linux, Linux is the kernel. A proprietary application, whether running on Linux or some other Operating System, could of course violate somebody’s patents. That has nothing at all to do with Linux.

Ah the old “Linux” means only the kernel argument. Really, the way that most people position Linux is that there is a Linux Kernel, and a Linux distribution. Linux can refer to either.

What is Microsoft? Microfibre Soft Toilet Tissue.

Kyocera Mita needs to license anything it probobly has something to do with Microsoft network file/printing sharing protocols, licensing fonts or Active Directory integration which has nothing to do with Linux. So what are these features and how will they equate into benefits to be consumer.

Touted as the “biggest advancement in fundamental transistor design in 40 years”, the new 45nm High-K process is what makes the new core possible. This technology allows for further CPU die shrinkage while reducing electrical leakage. This leakage is generally responsible for excess heat generation and power consumption, what is ultimately means for Intel processor is great transistor density, low transistor switch power, faster transistor switching speed, and overall better thermal and power efficiency. For the consumer, this will increase the speed of future processors, allow Intel to add new features, and lower cost due to a high density of chips per manufactured process wafer.

分类: 科技 标签:

Router, Switch and Hub

2007年11月12日 没有评论

Regardless of what impact the deal has triggered in the marketplace over the past year, ultimately it’s about meeting market requirements. Instead, they want to be able to better and more easily manage their IT assets no matter what hardware or OS platforms.

If you make it through the CCNA and are still interested in Cisco and internetworking, you’re headed down a path to certain success. To learn every bit of the material covered in this book, you’ll have to apply yourself regularly, and discipline. Try to set aside the same time every day to study, and select a comfortable and quiet place to do so. If you work hard, you will be surprised at how quickly you learn this material. Arrive early at the exam center so you can relax and review your study materials.

Since there are a bunch of different types of devices specified at the different layers of the OSI model, it’s also very important to understand the many types of cables and connectors used for connecting all those devices to a networks.

Hub, also called multi-port repeater, does not segment a network; they just connect network segments together. MAC address, known as a hardware address, is burned right into the network card.

Switches aren’t used to create internetworks, they do not break up broadcast domain by default. They’re employed to add functionality to a network LAN. The main purpose of a switch is to make a LAN work better, to optimize its performance, providing more bandwidth for the LAN’s users. And switches don’t forward packets to other networks as routers. Instead, they only “switch” frames from one port to another within the switched network.

Switches create separate collision domains but a single broadcast domain.

Collision Domain: an Ethernet term used to describe a network scenario wherein one particular device sends a packet on a network segments, forcing every other device on that same segment to pay attention to it. At the same time, a different device tries to transmit, leading to a collision, after which both devices must retransmit, one at time. Not very efficient! This situation is typically found in a hub environment where each host segment connects to a Hub that represents only one collision domain and only one broadcast domain. By contrast, each and every port on a switch represents its own collision domain.

It’s also likely that at same point you’ll have to break up one large network into a bunch of smaller ones because user response will have dwindled to a show crawl as the network grew and grew. And with all that growth, your LAN’s traffic congestion has reached epic proportions. The answer to this is breaking up a really big network into a number of smaller ones. Something called “networking segmentation”, you do this by using device like routers, switches and bridges.

Two advantage of using routes: They don’t forward broadcast by default; They can filter the network based on Layer 3. Router function: Packet Switching/Filtering/Internetwork communication/Path selection.

Remember their routers are really switches, they’re actually what we call layer 3 switches. Switch is layer 2, switch frames.

Router, not only break up broadcast domains for every LAN interface, it will break up collision domains as well. Each port on that switch breaks up collision domains, but it’s not all good – all devices are still in the same broadcast domain. Do you remember why this can be really bad thing? Because all devices must listen to all broadcast transmitted, that’s why. And if your broadcast domains are too large, the users have less bandwidth and are required to process more broadcasts, and the network response time will slow to a level that could cause office riots.

分类: 科技 标签:

Inventec replacement agreed

2007年11月8日 没有评论

Inventec agreed to replace two faulty platforms. I got the RMA number in this afternoon.

分类: 科技 标签:

Follow up with UT issue

2007年11月7日 没有评论

Zhu Feng and Lin Peng went to UT to update system with modified NFS scripts. The testing result in UT is quite different from those we got in our lab. When mount 15T volume, one node always keeps in 3s, and the other one takes about 10 seconds. If mount 11T,one node 0.04s, the other 2.5s. If mount 7.5T, one node 0.03s, the other 1.5s. For comparing, it takes about 2.5s on each node to mount 15T in our lab. Hence it seems according to UT environment, 7.5T would be the best choice as UT guys need mount operating as short as 1 second and they also want large volume(>10T). Jackie, the testing engineer in UT, did not accept the 2.5s mounting time. He told Zhu and Lin to re-install all the systems and recovery to default statues. Hence, these two guys have to go to UT tomorrow morning. Let’s wait for the result, and I think we also need to let UT know clearly, for their 2.4 kernel, our current version is the best resolution, though we will modify the latest version as 4.0.

分类: 科技 标签:

Install Windows Live Write

2007年11月5日 没有评论

The IE7.0 often crashes recently, so that I have to write in the notepad and then copy/paste the content to the browser windows. If so, sometimes IE still failed when I click “publish” button. This issue has annoyed me for a few days. I try to fix it by installing the latest blog editor: Windows Live Write, which is provided from Microsoft and seems to be dedicated for their Live space.

First, the below two packages are required: Windows Installer, Windows Update Agent

Second, install the packages, then execute Windows Live Write installation program. This process takes about 20 minutes in my laptop(T23), it would download some files from Microsoft Web site and install them later.

The 360safe software reported some programs want to modify the register table, that will add wextract_cleanup0 to run_once item. The “wextract_cleanup0” is a cleanup command used typically by Microsoft’s IExpress/WExtract self-extracting executables. Up until recently, typical Windows Update packages utilized the WExtract library (and still do for certain things — eg. MSN, WMP installations etc). When the SFX is run, it creates a temp setup folder, typically “IXP0000.tmp”, which then gets deleted automatically after installation completes. The runonce command you’ve posted is the command used to delete this folder. I accept this modification.

OK, the Windows Live Write is installed successfully, and it has nice interface and is easy to edit texts.

分类: 科技 标签:

The 11th China Public Security Expo

2007年11月4日 没有评论

The security market has very bright future, even KingSoft, JVC, which were decidated for Video software and hardware devices, both release DVR and remote networking monitoring resolutions.However, few companys provide the background storage, and they know nothing about it.It seems to be our good chance.

To Jose:
Message from Phillip at 14:23:11 Nov 3rd,2007
Jose, although you can dislike or not agree with Chinese culture and some mental actions, you should not critize them as bullshit in front of me and my colleagues. I think it is not polite and you words would hurt my heart that is always proud of my country. We have different opinions, but we should respect each other.

分类: 科技 标签:

UT issue is related to Kernel version

2007年11月1日 没有评论

rpc.gssd/rpc.idmapd/rpc.svcgssd are dedicated for NFSv4.
NLM: NFS Lock Manager

A quick glance tells that I don’t have rpc.lockd running just kernel lockd. Also “man lockd” says: The rpc.lockd program starts the NFS lock manager(NLM) on the kernels that don’t start it automatically. However, since most kernels do start it automatically, rpc.lockd is usually not required. Even so, running ti anyway is harmless.

In the morning, I commented “exportfs -r” in /etc/init.d/nfs, then it caused the NFS share can not be showmounted. It means this command must be executed before start up NFS daemon.

See how heavily each nfsd thread is being used, look at the file /proc/net/rpc/nfsd. The last ten numbers on the th line in that file indicate the number of seconds that thread usage was that percentage of the maximum allows.

The first number is the number of threads available for serving requests and the last eight are number fo seconds that each thread are 100% busy. If the last few numbers have accumulated a significiant amount of time, then your server probody needs some threads.

According to today’s testing for duplicating UT issue, if there’s only a nfs daemon starting ,the mount operatings are quickly in each nodes,even without “-o tcp”. However, more tests are needed to verify its performance when encounter heavy load. Perhaps just as what Gavin said, let UT update their kernel to 2.6.

By the way, it has been proven this mount issue is not related with the mount version. If implement mount2.12a in RH9(2.4) kernel, the dlm_sendd is still a grant CPU hog. Later, We use mount2.11y(UT environment) on kernel to mount, the mount operations are quicky and smoothly. Then, it seems to be caused by different Kernel verion between 2.4 and 2.6.

分类: 科技 标签: