Singapore Dedicated Server Bandwidth (Uplink) | Network latency | Environment monitoring
Xssist™ Group Pte Ltd Singapore Dedicated Servers Client Testimonials Blog Community Frequently Asked Questions Contact Page
Singapore Dedicated Servers
Control Panel System
Control Panel System
Xssist Blog

Uptime for dedicated server, VPS and shared server

It is true. Dedicated servers for single sites in general have better uptime compared to VPS, and VPS has better uptime than shared server.

The observation is as follows:

For shared servers, many sites are located on 1 physical server. All these sites compete for cpu, disk and bandwidth, and each site can be a backdoor to the server. There is usually no control over how each account uses the server resources, except where disk or bandwidth quotas are exceeded. Control panels for web hosting do not have CPU quotas; i.e. limiting cpu usage based on userid, or per process. There are so many applications installed, and of various versions. Don't expect the users to update their applications. Therefore, when a security hole is found, it is easy to gain access to several accounts on a shared server; usually through a php application, such as phpbb or phpnuke. With access to php on the accounts, the intruder (or "hacker") has as much access as if he has shell. Through this access, he can gain root privileges if the kernel is not up to date, or even if it is. Well, or he can just buy an account on the server. The motivation however, is usually to gain free usage of the server, to run psybnc, eggdrop, or scripts to scan other servers, or to udp flood their victims. The side effect is that some of these scripts take up 100% of the cpu, and leaves the other accounts starved of CPU. Or with the udp flooding, the server gets starved of bandwidth. Either way, the shared server gets starved of resources, and just can't keep up with legitimate requests to serve webpages. The server may be up, the network may be up, but your website is so slow, as to be unusable. Other consequences may be mass defacement, or malicious rm -rf of the entire server.

VPS, eg Virtuozzo, fares better, in the sense that each VPS has its own "contained" environment. Each VPS can be configured to have its own guaranteed bandwidth and CPU shares. The common bottleneck for VPS is disk I/O. Any VPS can use up the bulk of the disk I/O on a physical server. Vitualisation software needs to come with a means to control disk I/O allocation. This does not seem to be available currently. It can be controlled by giving each VPS its own dedicated harddisk; however the bus, whether SCSI or PCI, can still be hogged by a single VPS. Another issue which affects uptime for VPS involves SCSI. A failing disk for any VPS can bring down the physical server. The MPT (LSI Logic) driver does not seem to be able to recover from some SCSI errors, eg. "mptbase: ioc1: IOCStatus(0x004b): SCSI IOC Terminated", the driver will try a "task abort!", but server just hangs. The physical server for VPSes tend to be worked hard, from the CPU to the storage; and both hardware and software are not prefect. Servers that get worked hard tend to hit the bugs, and crash, leading to relatively lower uptime compared to dedicated servers. One advantage for Virtuozzo VPS: it is very easy to migrate the VPS from one physical server to another. Virtuozzo uses rsync to do the transfer; rsync the contents, shutdown the original VPS, rsync again to bring the new VPS up to date, bring up the new VPS.

Hosting on dedicated servers, given the reasons above, would have the best uptime. i.e. no other website or VPS to compete for resources. No other website which introduces security problems. Lower load on the CPU and storage of a dedicated server. Of course, this is vs hosting the same website on shared, or VPS. The dedicated server would still need to be of good quality. We are comparing like vs like; same quality of hardware. I will cover the types of dedicated servers that are offered by webhosting companies in another piece.

Lim Wee Cheong
11 March 2007

[Sysadmin] Access to servers via mobile device and ssh
[Sysadmin] RAID 0 scaling on SCSI U320, Bonnie++ 1.93c benchmark results
[Sysadmin] TODO (Apr 2007)
[Sysadmin] Recover from mistakes in /etc/fstab or e2label usage
[Sysadmin] Server overloaded?
[Sysadmin] Server load high: CPU bound
[Sysadmin] Smokeping: deluxe latency measurement tool
[Sysadmin] Smokeping
[Sysadmin] Jul 08 to Oct 08 updates
[Sysadmin] Weak link - downtimes caused by the organic being
[Sysadmin] BIOS upgrades - uniflash - hotflash
[Sysadmin] Sizing for Virtual Private Server (VPS) & SSDs
[Sysadmin] iphone, ipod - bluetooth keyboard - Nokia e51
[Sysadmin] e2label, fdisk, /etc/fstab, mount, linux rescue, rescue disk, CentOS
[Sysadmin] opensuse, fix waiting for mandatory device, eth0, eth1, eth2, eth3
[Sysadmin] mount: could not find filesystem '/dev/root'
[Sysadmin] Parallels Virtuozzo Physical server to Container migration (vzp2v)
[Web hosting] DDOS (Distributed Denial of Service)
[Web hosting] Uptime for dedicated server, VPS and shared server
[Web hosting] Shared, Guaranteed and Dedicated Bandwidth
[Web hosting] Unmetered bandwidth
[Web hosting] Free domains?
[Web hosting] Joomla Scalability
[SPAM handling] Tracking applications which are exploited for mass spam mailing
[Buzzwords] Clusters, Clustering
[Security] Destruction of faulty hard disks
[Storage] Benchmark using iometer on linux
[SSD] Benchmark Intel X25-E and Intel X25-M flash SSDs
[SSD] Intel X25-E 64GB G1, 4KB Random IOPS, iometer benchmark
[SSD] Intel X25-M 160GB G2, 4KB Random IOPS, iometer benchmark
[SSD] Comparison of Intel X25-E G1 vs Intel X25-M G2
[cPanel] ClamAV version has reached End of Life! Please upgrade to version 0.95
[cPanel] How to install Java, ImageMagick and ffmpeg
[Perl] Opening text files for reading, and simple regexp (regular expressions)