E2E Networks » performance

Independence month discounts

tarun — Mon, 23 Aug 2010 19:10:45 +0000

For immediate release :-
We an India based Hosting and Cloud Computing start-up, aim to help all our peers with equal ambitions and energy to strive and succeed. Hitherto it was only the large media houses and venture backed startups that could do snappy websites using expensive low latency datacenter facilities in India. We have pulled the powers from large media houses, corporates and venture backed start-ups, by designing special ‘low latency’ hosting services for start-ups at highly competitive rates.

We know long loading times have subtle effects on user behavior as corroborated by many other studies. If you are a start-up competing with big guys hosted locally you know you need to equal or better their snappiness advantages and if your competition is hosted in US then *move to India before your competition does to create an edge*

For technical understanding on why latency matters please refer the blog and make a well informed decision while choosing hosting for your servers in India instead of high latency locations like US if your web visitors are from India.

As a special Independence month offer we’ll waive off 50% of the onetime setup fee on all dedicated server hosting in India if you sign up on or before EOD 31st of Aug 2010. Also avail 5% discount on usage of coupon code [ India-Aug-2010 ] on monthly fees on all our dedicated server hosting in India and Virtual Private Server hosting in India, over and above the other automatic discounts mentioned on our website

For startups who would like to start small please check out our India based VPS server hosting plans starting from Rs. 1799/mo . We offer sane support for even un-managed India based server hosting plans with a lot of free security and scaling advice on application architecture side.

For serious bloggers using E2E Networks VPS servers we provide generous on-demand capacity bursting capabilities for free if you give us a heads up on a breaking news story that could get dugg/slashdotted an hour or two in advance.

Your feedback on how we can bring hosting back to India is important to us and we welcome it at all times just as we welcome traceroutes from everywhere in India to us, regardless of whether or not you are an existing customer of E2E Networks.

It will happen, it will happen in our lifetimes

tarun — Sun, 15 Aug 2010 07:03:19 +0000

India’s first barcamp was practically a turning point for the startup scene in the country or atleast it appears so to yours truly.
Slideshare.com was practically ideated by participating uzanto.com folks at the venue. Geobeats was looking for technology driven co-founders here. It was truly exciting to be amongst all these folks and many more

The presentation by Kunal whose upshot is this slide had the maximum impact on what ails the Indian broadband penetration numbers, the idea that different companies can fix the different parts of this equation was clearer, before this I always thought a wireless ISP at my hometown would be a nice upgrade to its infrastructure :-

So here we are at E2E Networks Private Limited reiterating our promise ‘It will happen, it will happen in our lifetimes’
Pervasive and world’s fastest broadband Internet in every Indian home that can afford a Rs. 750 or more a month connection and this would happen only if everyone who has to serve Indian audience hosts in India. We are doing our part of the bargain which is to bringing Internet traffic ( dependent on content and applications ) back to India so your last mile ISPs can stop complaining about the high cost of International Bandwidth, the cost of National long distance backhaul is a no-brainer there is million kilometers of practically un-used dark fiber capacity in the country today, the last mile is already available to the extent of 8Mbps ( no wonder you have those IPTV and other anti-net neutrality products riding the last mile ). So whether your Internet business uses a mere 100GB bandwidth a month or several 100 terabytes we are are ready to serve your hosting needs. Its not just about performance, bringing hosting back to India also means your contribution to growth of broadband in the country and consequently to the growth of your own marketplace.

So in the enlightened self interest of helping grow the broadband marketplace we are launching the phase 3 of bandwidth price war in a few more weeks from now the E2E Networks’s Cloud Computing Products cheaper than anyone else in the Asia Pacific region.

And oh by the way we’ll waive off 50% setup fee till the end of August starting today for next 15 days. The signup must be completed before 31st August 2010 to avail this special offer. So bring your servers back to India.

Why Latency matters in India?

tarun — Tue, 04 May 2010 09:06:40 +0000

Loading…Frustrated on seeing this? The large visible delays in loading a website are really irritating. In this fast paced world, your web visitors don’t want to wait for long for your website to load. A few major causes of latency are discussed here.

One of the main reason for latency is the physical distance of the datacenter where the site is hosted from your visitors. If you are hosting your website on servers in one geographic location but you are receiving visitors from another geographic location, the latency would be high. The sites hosted in countries like US/UK add up a latency of about 300ms to 400 ms which increases page rendering times by a few seconds. With larger number of page elements higher latency makes for higher page load times regardless of optimizations in the size of images and compression of javascript/css elements.

As an example of the effects of latency lets look at a few data points :-

Ping a website hosted in India ( say pluggd.in ) and another hosted in US (mobikwik.com)

Pinging www.mobikwik.com [67.18.92.243] with 32 bytes of data:

Reply from 67.18.92.243: bytes=32 time=344ms TTL=50

Pinging pluggd.in [180.151.0.203] with 32 bytes of data:

Reply from 180.151.0.203: bytes=32 time=53ms TTL=49

The huge difference in latency times of both the websites is clearly visible.

Firebug is an extension for Mozilla FireFox, and can also be used to analyze your site’s network activity.

The firebugs output for both the websites could be seen at

http://e2enetworks.com/wp-content/uploads/mobi.png

http://e2enetworks.com/wp-content/uploads/pluggdin.png

See the net effect in being able to throw in 470KBytes of data in 12.72 seconds mostly from India compared to being able to throw 171 KB of data at the browser in 13.74 seconds. Hosted in US no matter how much processing power you throw at something you hit a lower bound for each content element being pulled in 600 ms or more.

The firebug output for a nearly pure everything in India kind of site can be seen here

http://e2enetworks.com/wp-content/uploads/tarundua.png

There are several hosting companies in India which have their servers in USA and thus sites hosted by these get impacted by high latencies. How to determine if your site is in USA or India.

Low latency means faster browsing experience and that’s what every end user wants but also higher page views and greater engagement as shown by Google’s study.

Is your webhost really in India ? How it affects your site speed

tarun — Mon, 30 Nov 2009 14:34:29 +0000

Is your website target audience in India ? If 70% or more of your web traffic is from India this for you.

Why should it matter ?
If your site is hosted in US have you noticed the long delay before your webpages start appearing on the browser. While some of the other big media sites are ‘snappy’ as in they start instantly loading up. This is because of the ‘latency’ due to physical distance from the datacenter where the site is hosted.

There is a lot of performance optimization that you can do with CSS sprites ( reduces the number of image elements to load) , combining CSS files, javascript minify, concat and at the bottom apart from other Y! Slow rules.

Dedicated RAM, sufficient IOPS and fair share of CPU are some other elements that need to be in place so a web visitor to your site is not queued for several seconds before she gets her first byte of your website.

Add to it hosting in India at < 85ms latency, it gets you another level of performance boost beyond these optimizations where you can aim for snappiness that only big boys in the media industry could hope to buy till now in India. Think a 200 ms per page element saving on load times. Google's research shows it is important to reduce latency of your webpages http://code.google.com/speed/files/delayexp.pdf

A lot of Internet players in India now recognize the importance of having their websites available at a lower latency for faster response times for the end users. The latency to US/German datacenters is around 300/270 ms which translates into several additional seconds while loading up pages of websites. The big players have always chosen to host locally in India despite the high costs long associated with hosting in India.

How do I find out if my datacenter/webhosting company hosts me in India
Assuming you are in India and if you are on windows use the run dialog to open up the command prompt by typing in ‘cmd’
ping
is your friend. As an example :-

C:\Documents and Settings\Tarun Dua>ping e2enetworks.com

Pinging e2enetworks.com [180.151.0.212] with 32 bytes of data:

Reply from 180.151.0.212: bytes=32 time=81ms TTL=54
Reply from 180.151.0.212: bytes=32 time=82ms TTL=54
Reply from 180.151.0.212: bytes=32 time=81ms TTL=54
Reply from 180.151.0.212: bytes=32 time=81ms TTL=54

Ping statistics for 180.151.0.212:
Packets: Sent = 4, Received = 4, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
Minimum = 81ms, Maximum = 82ms, Average = 51ms
Less than 100ms (milliseconds) ping round trip time is a good indicator if your webhost is located in India. More than 280 milliseconds is fairly a UK or US based host who remotely manages servers from India.

So I would really like to move to an Indian web host ?
Check out our hosting plans for dedicated servers and Virtual Private Servers

Demystifying latency in webpage requests

tarun — Tue, 21 Jul 2009 07:16:49 +0000

Since the advent of the Internet and the World Wide Web (www) in the last century, till today there have been trillions and trillions of Internet page requests made across the world. However, webmasters feel all the more curious to know how actually a webpage request is being served on their user’s computer and to use that understanding to try to speed it up.

Data on the Internet is packaged and transported across in small data packets. The regular or irregular flow of these data packets affect the user’s Internet experience. Whenever, one sees a continuous flow of data on his screen, this in turn means that the data packets are moving across smoothly and in a timely fashion. However, if the same data packets move across with large and visible delays, that means that the user’s experience is degraded and he may feel frustrated at the poor network connection and speed.

In this article, I seek to demystify all such beliefs and bring parity to our understanding of webpage request. I shall make an attempt creating a better understanding of latency effects of network and low bandwidth on a webpage request.

Lets first develop an understanding of some of the networking concepts to build our follow up understanding of the webpage request and latency. With the advent of networking, it was thought that millions and millions of users would be connected through a common network, hence the idea of TCP / IP model.

The key features of the TCP/IP model is encapsualtion, which is the concept of collecting the data and covering it with a common container for transmission. The common container is called the “IP Datagram”, also known as “IP Packet” or just the “Packet”. This IP Packet is a simple thing, with a header which contains the information used for routing the the packet to the destination and followed by data which is any information sent across.

Lets now concentrate all our energies on understanding another important concept in networking, the OSI model. It was created to lay out the process of turning the application data into something that can be transported through the Internet. The upper layers of the OSI model describe as to what happens within the applications running on the computer. These include the human-machine interface, conversion of the high level language into machine language, encryption, authentication and permissions. The lower layers are the ones where to and from applications are turned into data to move across the network. This is where data encapsulation occurs and the IP Datagram or “packet” is built.

The transport of data across the network is a 3 step process:

1. Data from the source is passed through the TCP/IP stack and wrapped into IP Datagrams, commonly known as “Packets”. These packets are then transmitted by the source computer in the network

2. Packets are passed along the network until they reach the destination computer

3. Packets are received by the destination computer and are passed through the stack

According to Wikipedia, Latency is the time delay between the moment something is initiated, and the moment one of its effects begins or becomes detectable. The most common understanding of Latency is it takes time for web pages to load and for emails to reach the destination inbox. Though, this is a form of latency, however, lets take down latency as the time delay imparted by each element involved in the transmission of data. Lets develop our understanding of what causes latency. The are many logical and physical elements involved in networking.

Application Latency

The need to read and write to disk causes some time delays. The processor could be very strong and highly rated, however, it still has limitations to as to what it can read and write in stipulated time. It takes a finite amount of time to manufacture data and present it. There are a lot of hardware limitations as well, such as the amount of memory which affects application performance.

Serialization Latency

The encapsulation of data (as discussed above) is called serialization and takes a finite amount of time. It is calculated as follows: Serialization Delay = Packet Size in bits / Transmission Rate in bits per second Serialization can lead to significant delays and latency on link that operate on low transmission rates.

Routing & Switching Latency

A network request causes data to flow from point A to point B. This would be simple, if the network was just 2 computers, however, this is not to be. In networks like the Internet, data and hence the packets are transmitted from source to destination through a series of routers and switches connected through circuits, which are hardware devices needed for transmission through the network.

These hardware machines have to manage the Internet traffic causing delays caused by the routing and switching process. This refers to the amount of processing time for a router or switch to receive a packet, process it and transmit it.

These days, with the advancements in the computer hardware technology, these delays have reduced to only a few nanoseconds. High performance routers and switches each usually add upto 200µs of latency to the link.

Queuing Latency

Queuing latency refers to the amount of time a packet spends sitting in a queue waiting for transmission due to over utilization of the link. Though over-utilization of high speed Internet backbone is very rare, but it can be easily seen on lower speed networks. Congestion can cause these delays to become infinite since packets may be dropped when router becomes full. Routers use various queueing management algorights to ensure latency is minimized. The most commonly used WRED algorithms bound queueing latency at 20 ms.

Propagation Latency

Propagation latency is the delay caused by the transmission medium. The amount of slowing down is known as the Velocity Factor (VF). Typically, there are 3 medium of transmission of data across the networks, copper cables having a VF in the range of 40% – 80% of the speed of light, fibre-optic cables leading to a VF of around 70% of the speed of light and the electro-magnetic radio waves having the least possible VF. This delay happens even without considering the amount of data being transferred, the transmission rate, the protocol being used or the link problems.

Transmission Rate and Bandwidth Latency

Transmission Rate is the term used to define the number of bits that can be extracted from the medium. It is commonly measured in the terms of number of bits per second. The maximum transmission rate defines the fundamental limitation of the transmission medium. Generally, Copper links have a maximum transmission rate of 10, 100 or 1000 Mbps. For Fibre-optic links, transmission rates vary from around 50 Mbps to 10 Gbps.

Wireless LANs and satellite links use a modem to convert the bits into a modulated wave and then on transmission convert them back into bits using the demodulator. The limiting factor in these type of links is the limited bandwidth available to these signals. The amount of radio spectrum occupied by any given signal is called its bandwidth. Since radio spectrum is a limited resource, the occupied radio bandwidth is an important limiting factor in wireless and satellite links.

Protocol Latency

Lets now take a look at the network data exchanges. Connectionless data exchange is the one where data is pushed through with any consideration. Here the packet traverses the Internet to search for its destination, however, if something happens to it midway, nothing can be done. This is usually used for streaming music, videos, VOIP. The protocol used is User Datagram Protocol (UDP). It doesn’t have any overhead or connection management. There is no retransmission of data as well.

On the other hand are the connection based data exchanges. They rely on the establishment of the connection which manages every packet that is transmitted. The transport protocol used is the Transmission Control Protocol (TCP). It provides for the error free delivery of packets and hence the data. TCP connections have 3 phases:

1. Establish the connection

2. Send the data

3. Close the connection

All this adds to the time being taken while the data is transmitted and hence the delay and Latency.

This puts the webpage request on the table and opens it thread bare to clear the air on what goes behind each of our clicks while we are on the Internet connected to the millions or billions of users and trillions of data. We make an understanding of the time delays or Latency and now agree that it is imperative and necessary.

References

1. What is network latency? and Why does it matter? -

http://www.o3bnetworks.com/docs/O3b_latency_white_paper2.pdf

2. Satellite Internet Access - http://www.sisp.net/broadband/satellite.htm

3. Network bandwidth and Latency -

http://compnetworking.about.com/od/speedtests/a/network_latency.htm

4. Anatomy of a HTTP Request -

http://www.websiteoptimization.com/secrets/metrics/10-21-http-request.html

Improving linux IO performance

tarun — Tue, 28 Oct 2008 15:17:08 +0000

1. Mount options: use noatime

Most Linux server machines can do without last access time modified for every file and each directory which is being read. So I’ll just go ahead and re-quote for the nth time what Linux Kernel developer Ingor Molar has to say to emphasize the point.

i cannot over-emphasise how much of a deal it is in practice. Atime
updates are by far the biggest IO performance deficiency that Linux has
today. Getting rid of atime updates would give us more everyday Linux
performance than all the pagecache speedups of the past 10 years,
_combined_.

You can simply remount your filesystems without rebooting your machine using remount option.
As an example:-
/bin/mount -t ext3 -o noatime ext3 /dev/sda5 /
for remounting
/bin/mount -t ext3 -o noatime,remount ext3 /dev/sda5 /

And don’t forget to modify corresponding lines in your /etc/fstab
/dev/sda5 / ext3 noatime 1 1

2. Use tmpfs
Speedup heavy read-write IO for temporary data stores by by using memory instead of disk.

3. On systems not constrained for memory reduce swappiness of the Linux machine
/bin/echo “10″ > /proc/sys/vm/swappiness

4. Set blockdev readahead to a reasonable value to improve read performance
/sbin/blockdev –setra 131072 /dev/sda

The default readahead value is too small.

Linux: In memory filesystems tmpfs vs ramdisk

tarun — Sat, 25 Oct 2008 07:37:10 +0000

Although reading and writing to files is fast in Linux with aggressive readahead and caching, it can still slow down applications that make extensive use of ondisk temporary files. As an example MySQL which can do a lot of on disk temporary tables if the temporary tables need to have a large varchar, text or binary column.
It makes sense to mount an in-memory filesystem on the MySQL’s tmpdir usually /tmp to ensure that your on-disk temporary tables are rapidly written to and read from memory to return query output fast by avoiding expensive disk IO. Similarly a lot of different web applications can derive a lot of benefit by writing temporary data to an in-memory filesystem as opposed to the disk.
The two choices ramdisk and tmpfs
Linux Kernel loads up 16 Ramdisks of 16 MB each at bootup time. They don’t occupy any memory space at initialization. Ramdisks allocate memory when they are put to use by formatting them as ext2 or some other non-journaling filesystem. ( no not ext3, there is no use of journaling for an filesystem that is transient ) Once allocated the memory can’t be returned from a ramdisk to the operating system. Ramdisks suffer from another limitation that its size can’t be dynamically increased.
tmpfs on the other hand doesn’t need to be formatted as another disk filesystem. Its can be dynamically resized. Un-utilized memory can be used by the operating system. The only downside is that tmpfs can also use Virtual Memory and its contents can be swapped out causing disk IO that we seek to avoid by using tmpfs. However swappiness should be minimized on a mission critical server anyway by tuning the /proc/sys/vm/swappiness value at boot time.

As an example of how to use tmpfs
/bin/mount -t tmpfs -o size=1G,nr_inodes=10k,mode=0775,noatime,nodiratime tmpfs /tmp

to dynamically increase its size
/bin/mount -t tmpfs -o size=2G,nr_inodes=20k,mode=0775,noatime,nodiratime,remount tmpfs /tmp

Tuning the swappiness
/bin/echo “1″ > /proc/sys/vm/swappiness

These need to be added to /etc/rc.local to make the settings persistent across reboots.