Tag Info

Hot answers tagged

7

Use a benchmark tool like bonnie(++). It's easy to install on about every distribution, and since it measures different aspects, you get quite a good picture how the system performs in a given situation. If you just want to use basic tools, you could use dd: For write speed: dd if=/dev/zero of=outputfile bs=512 count=32M (The product of bs and count ...


7

I had a similar problem trying to raise the speed of a drbd synchronization over two gigabit links some time ago. In the end I managed to get about 150MB/sec synch speed. These were the settings that I applied on both nodes: ifconfig bond0 mtu 9000 ifconfig bond0 txqueuelen 10000 echo 3000 > /proc/sys/net/core/netdev_max_backlog You could also try to ...


6

Probably it's the encryption. You can try scp with different ciphers, for example: scp -c arcfour src dest Check ssh_config manual page for available Ciphers. RC4 (arcfour) is a fast cipher, but probably not as secure as some alternatives.


6

The three commands and ZFS setups you've listed are vastly different configurations. zpool create tank mirror disk1 disk2 disk3 disk4 This creates a 4-way mirror with the capacity of ONE disk. Lots of protection, less space and less performance. I don't think you want that. zpool create tank mirror disk1 disk2 mirror disk3 disk4 This creates a stripe ...


5

Part of it may have to do with the internal implementation of OpenSSH. Take a look at HPN-SSH for a decent explanation for one reason for the bottlenecks, and patches to OpenSSH that can resolve it (if you're willing to patch and rebuild from source).


5

The mysql command show status like 'Queries' gives the total number of queries. You can periodically run a script which grabs this value and calculates the query rate for a particular time interval using the difference. ie If you run the script every minute and get the change in total queries for each run and divide by 60 you will have a rough estimate of ...


5

few things: did you try adjusting mtu to make use of jumbo-frames? are you absolutely sure that the link between the two servers does not have any packet losses? does ethtool show you any errors on the interfaces on both ends? what does top/atop say during the prolonged test - do you see any of the cores fully occupied by iowait? you'll [most probably] ...


4

iperf would be one way. You could design a bunch of tests and manually and then run iperf under each scenario. However, have a look here. This is how these guys do it. http://www.smallnetbuilder.com/lanwan/lanwan-howto/26563-how-we-test-hardware-routers- They use a tool call lxChariot which is meant to simulate real world loads.


4

Evan, Based on your response "the interface on the iMacs and laptops will be gigabit (theoretical)" - then, as Zoredache pointed out - your theoretical max is 120MB/s. It'll be lower, as Chris S said, allowing for normal network overhead. The system is as fast as the slowest NIC. In this instance, that's the gigabit NICs on the workstation. You also ...


4

I'm trying to calculate the bandwidth delay product between various hosts... Have I missed something obvious, or is the article wrong? You're only missing the units; when you multiply bps by seconds, you get bits as the units: 10Mbps * .029s = 2900000.0 bits (362500 bytes) The point of the article is that it highlights what your (scaled) TCP window ...


4

For the most part, you can ignore lo traffic. That is your loopback adapter, which is used by various linux subsystems. That adapter exists purely within the linux kernel, and traffic through it does not touch your eth0 interface. The connections you're seeing from high ports are expected. Those are the ports the client opens to make connections to the ...


3

Your theoretical maximum throughput at 1Mb/s would be 60 Mb/min, or 7.5 MB/min; (remember 8 bits in a byte) 3600 Mb/hour, or 450 MB/hour Note that this is highly theoretical and does not include things like ethernet, IP, or TCP headers and checksums; non-data carrying packets such as SYN, ACK, various ICMP packets any allowance for non-duplexed data ...


3

The general perception is that Cisco doesn't lie about their specs. Having said that, they probably aren't real world specs. On the firewall, there is so much to configure that ANY inspection you do on a packet is going to rob you of throughput. Add the AIP-SSC card for IPS inspection and you're going to bury your firewall. With your rough figures, I ...


3

Start simple and iterate. Trunk them together with single GigE. If you've got ~20 workstations and printers in a typical office, you're probably not hitting GigE wire speed at this point anyway. Next step is to do GigE link aggregation (802.3ad). Try 2 GigE connections between each switch, or maybe 4. If your traffic is pretty evenly spread over your ...


3

Take a look to postgres stat tables. To get total number of commits, use query like SELECT SUM(xact_commit) FROM pg_stat_database - this query would return total number of successful commits on all databases since server start. Running this query every N minutes/seconds would allow you to draw nice graph (subtracting previous value from the current one ...


3

I'd suspect your PCI-e x4 slots are the bottleneck. The theoretical throughput on those slots should be in the range of 16 Gbps (saturating the NIC with room to spare), but that's not always well-implemented from the controller standpoint. Got an x8 or higher slot you can steal from something else to test?


3

It means absolutely nothing. Your provider can only guarantee a certain bandwidth to the edge of their network -- it's entirely possible that network congestion is going to cause problems beyond their bounday. Also, an upload to Amazon is likely to be using a single TCP connection, and the bandwidth-delay product could very well be biting you in the butt.


3

Most importantly, for best performance you need 24 drives (or more, bear with me), because you have 24 threads. If there are less disks than threads, you do not have sequential operation. Considering two threads on a single disk, it will have some seeks; each seek is say 10 ms, so a loss of about 1 MB of transfer opportunity. With only 10 seeks per second, ...


3

Yes, this is also a known attack on firewalls. The reason for this is that routers and firewalls can process a number of amount of packets (their size doesn't matter). So to calculate the bandwidth they choose a certain packet size and multiply it by the amount of packets they can process. For instance if they calculate that 1 gbit is calculated with 512 ...


3

1.) Pulling a file down with wget doesn't necessarily accurately measure the speed of your connection as it's subject to upstream traffic conditions on whichever of many paths your provider may be using to reach your target host. You're additionally also measuring the performance of the TCP setup on your OS and even the efficiency of wget itself. This sort ...


3

Adding the noac nfs mount option in fstab was the silver bullet. The total throughput has not changed and is still around 100 MB/s, but my read and writes are much more balanced now, which I have to imagine will bode well for Postgres and other applications. You can see I marked the various "block" sizes I used when testing, i.e. the rsize/wsize buffer ...


3

What you are seeing is not a time out. What you are seeing is the result of using the wrong protocol for the purpose. TCP will perform flow control, which means it will adjust the transmission speed to the capacity of network and receiver. Additionally TCP will retransmit lost packets. UDP does neither of those. The nc command you used will transmit ...


2

Try disabling checksum offloading under System > Advanced. That functionality on your NIC may be broken, or the driver may not work properly with it on that NIC for some reason. That would be my first guess, given the NIC change, and the fact that you aren't exactly using the best NICs on earth. Another problem with similar symptoms could be some sort of ...


2

Going with what Josh said, only one NIC should come into question when you're running links locally from the pfSense box. Are you sure that the Realtek 8169s are assigned to your WAN interfaces? If you hit up the console and drop to a shell you can run ifconfig to be sure (the Realtek NICs should be re0 re1, etc.).


2

I strongly suspect you're seeing a bandwidth-delay product (BDP) artifact. A "long, fat pipe" (high bandwidth, high latency) limits the amount of data that TCP can put "in flight" at any given time. Your observation that parallel transfers run faster is a big indicator of BDP coming into play. What's your round trip latency to S3 look like? You might try ...


2

The bandwidth of your connection is going to be the primary determining factor on the rate of data transfer. Otherwise, you could do some comparative testing. If you mean some limit on the total, I don't think there is one. What is it that you're trying to do? What problem are you trying to solve?


2

You can try iperf, it works as a client and server so you need to install it on two computers, here's how you use it run on Server iperf -s run on client iperf -c <ip of server> hope that helps, RayQuang


2

You could use a tool such as vnstat to graph your bandwidth. Once vnstat is running, launch a big copy of a file from a computer to another, using dd for example (try to really limit the CPU usage so it doesn't get to be the limiting factor). If your HDD is fast enough (so the network is the limiting factor), it should give you a good idea of your bandwidth. ...


2

What about two NICs on the webserver, one on the DMZ and one on the LAN? Edit: Since the answer was accepted I am putting more details. The webserver is necessarily public facing, the idea is to firewall that so only ports 80,443 are publicly accessible. Then internally it can communicate with the database server on a LAN interface. This also has the ...


2

I've been banging my head against this problem all day, with no solution found. You can turn up the debugging of the NFS server, but that doesn't provide much detail (if that exmaple is accurate) and will probably dominate a busy NFS server's disk with useless baggage logged in addition to the bare filenames. Another solution is adding rules to ...



Only top voted, non community-wiki answers of a minimum length are eligible