Short answer. Before replacing a slow server, five checks, in this order. The hours: if it is slow at fixed times, look for what runs at those times, backup, antivirus, reports. The four resources, read at the moment it crawls: processor, memory, disk, network; one saturated resource is enough. The database: its size, its upkeep, and the limits of its edition. The rest of the chain: if the branch is slow and head office is not, the server is not to blame. The system: a version leaving support has to be replaced anyway; Windows Server 2016 reaches that point on 12 January 2027. In this guide's example, a few settings and then two parts bring forty seconds down to three.
This guide is for the manager of a company whose management software, accounting or files run on a server at the office, for anyone being told to replace it, and for the founder based in France who hears on the phone that “it is crawling” without being able to look. The yardsticks quoted come from public sources: the end-of-support dates published by Microsoft, the limits of SQL Server Express, the orders of magnitude of disk operations per second. The numbers in the figures are a constructed example: an eight-year-old server, twenty-two workstations, management software and its database. This guide gives no hardware prices.
The server's day: forty seconds at 10:30, three the rest of the time
The figure overlays what the server does hour by hour and the time it takes to save a delivery note.
What the day shows. The team works from eight to five. The backup starts at half past nine, because someone set it that way years ago; the antivirus launches its full scan at ten; at the same moment accounting runs a report that rereads three years of invoices. Between half past nine and half past eleven the three overlap, and the saving time goes from three seconds to fourteen, thirty-eight, forty, then twenty-two. For the rest of the day the same server answers in three seconds.
The first setting. Move the backup and the antivirus scan to after closing time, without overlapping, and produce the heavy report at night, as described in the guide to background tasks. For the antivirus, Microsoft recommends excluding the database files and the program that runs the database from scanning: it improves performance and avoids files being locked when the database needs them; the trade-off, which Microsoft also points out, is that an excluded file is no longer watched. In the example, these settings bring the worst moment from forty seconds down to twelve, with nothing bought. What to note in your company. The hours when it crawls, over one week, and the list of tasks scheduled on the server: if the two coincide, you have the first cause.
«The server is not slow all day: it is slow when everything is asked of it at the same time.»
Four resources at 10:15: it is not the processor
The figure reads the server's four resources at two moments of the same day.
What the gauges say. At a quarter past ten the processor is at 38 %: it is waiting. Memory is at 97 %: the system has none left and starts using the disk in its place. References describe this: when real memory is overcommitted, performance degrades or even collapses. The disk is busy 100 % of the time, and each operation takes 180 milliseconds instead of 9. The network, at 9 %, is not involved. A faster processor would have changed nothing; yet that is what gets bought when a server is replaced without measuring.
Why the disk decides. A mechanical hard disk at 7,200 revolutions per minute performs around 75 random operations per second; SSDs perform thousands, and the most recent ones hundreds of thousands. A database mostly does random operations: as soon as it no longer fits in memory, everything rests on that figure. The limit people forget. Much management software ships with the free edition of its database engine. For SQL Server, the Express edition is limited to 10 GB per database, 1,410 MB of memory for its data and four processor cores. Two consequences: adding memory to the server does not speed up a database locked in that edition, and a 9.4 GB database, as in the example, will stop growing at 10. The answer is to archive old years, or to move to an edition that fits the size of the company.
«A faster processor would have changed nothing; yet that is what gets bought when a server is replaced without measuring.»
Six common causes, and how to recognise them
Each row gives the sign that betrays the cause, and what is done.
Heavy tasks in the middle of the day: slow at fixed hours; that is the first figure. Not enough memory: memory at the ceiling and a disk that works non-stop, especially when remote desktop sessions run on the same machine as the database. Mechanical hard disks: the disk at 100 % while the processor waits; two mirrored SSDs for the system and the database change more than a new processor. A disk array with one failed disk also slows everything else down until it is replaced: that is written in the server's log, which nobody reads.
A database that has grown: slowness sets in year after year, because searches reread longer and longer tables; the answer lies in indexes, as explained in the guide to the slow app, and in archiving old financial years. Too many roles on one machine: files, database, remote desktops and sometimes mail on the same server; each gets in the others' way. The server is not to blame: slow at the branch and fast at head office, it is the link, as in the guide to linking sites; slow on a single workstation, it is the workstation; slow for everyone on Wi-Fi, it is the network, as in the guide to network installation.
Six checks before buying a server
The second is accented: the list of scheduled tasks is the check that avoids the most purchases.
The hours: when it crawls, for whom, on which screens. The tasks: everything scheduled on the server, with its time and duration. The four gauges: processor, memory, disk and network, read continuously for a week, and looked at for the hours noted.
The database: its size, its growth per year, its edition, the date of its last maintenance. The system: its version and its end-of-support date. According to the dates published by Microsoft, Windows Server 2012 R2 left extended support on 10 October 2023, Windows Server 2016 leaves it on 12 January 2027, Windows Server 2019 on 9 January 2029; SQL Server 2016 left it on 14 July 2026. A system that no longer receives fixes is an open door, as the guide to ransomware recalls: it gets replaced, even if it is fast. The decision: tune, reinforce or replace. A recent machine, under warranty, with a supported system, is reinforced; an eight-year-old machine whose system is leaving support is replaced, and that is the time to decide where what comes next will live, at the office or with a host, as in the guide to hosting.
A server that responds, in five layers
The accented layer is scheduling: it is the setting that requires no purchase.
Hardware. Memory for what the machine really does, mirrored SSDs for the system and the database, a UPS, and a warranty that is still valid. System. A supported version, updates applied in the evening, and one role per machine or per virtual machine, so that one task does not get in the way of the others. Data. A maintained database whose size is tracked and whose edition fits its use.
Scheduling. Backups, scans and heavy reports outside working hours, without overlap, with a backup whose restore is tried, as in the guide to backups. Measurement. A permanent reading of the four resources and an alert on overrun: you know why it is slow before anyone calls. It is part of the routine described in the guide to managed IT. And the software itself. When the server is healthy and the software still crawls, it is the software: screens that reread the whole database every time they open, an old engine, an architecture from another era; the guide to migrating old software describes what follows.
Getting a fast server back in 2027, in weeks
Three weeks in the example, the first with nothing changed.
- What the manager sees: the week's curve, the slow hours and their cause, the size of the database, the system's end-of-support date, and a written recommendation: tune, reinforce or replace.
- What the company no longer does: wait forty seconds every morning, buy a server without knowing what was missing, discover that the database has hit its limit.
- What is measured: time to save a document at peak hours, memory and disk at those hours, time and duration of scheduled tasks, size of the database.
- What Symloop does: measure for a week, tune what can be tuned, reinforce or replace with a written file, migrate without stopping the business, then monitor; and when the slowness comes from the software, take it over or rebuild it.
Week 1, measure: the four resources read for a week, slow hours noted by the users, scheduled tasks listed, the size and edition of the database. Week 2, tune: heavy tasks moved to the evening, antivirus exclusions on the database, indexes, old years archived, overdue updates. In the example, the worst moment goes from forty seconds to twelve.
Week 3, reinforce or replace: memory and two mirrored SSDs if the machine is worth it, and the time falls to three seconds all day; otherwise a new server or hosting, with a migration prepared, rehearsed on a copy, then done one evening. In the example the machine is eight years old and its system leaves support in January 2027: tune straight away, reinforce to the minimum, and prepare the replacement for the first quarter, instead of buying in a hurry a machine that would have kept the same backup at half past nine.
- Why is our server slow only in the morning?
- Because something other than your users is asking it for work at that moment: most often a backup, a full antivirus scan, updates or a heavy report, scheduled in the middle of the day. The server's list of scheduled tasks, compared with the hours when it crawls, gives the answer in a few minutes.
- Should a slow server be replaced?
- Not before measuring it. If it is slow at fixed hours, it is a scheduling matter. If it lacks memory or runs on mechanical disks, it can be reinforced, provided it is still under warranty and on a supported system. It is replaced when it is old, when its system is leaving support, or when reinforcing it would amount to rebuilding it.
- How do I find out what is slowing the server down?
- By reading four resources continuously for a week: processor, memory, disk and network, and looking at their values at the hours when users complain. The one at the ceiling at those hours is the cause. A reading taken in the evening, when everything is quiet, shows nothing.
- Is adding memory enough?
- It often helps, especially when remote desktop sessions share the machine with the database. But check the edition of the database engine: SQL Server Express uses no more than 1,410 MB for its data, whatever memory the server has. In that case it is the edition or the disk that has to change.
- Does an SSD really change anything?
- For a database, yes. A mechanical hard disk at 7,200 rpm performs around 75 random operations per second; an SSD performs thousands or more. As long as the database does not fit in memory, that figure sets the response time. They are fitted in pairs, mirrored.
- Our database is nearing 10 GB: what happens?
- If it runs on SQL Server Express, it cannot exceed 10 GB: at that size it can no longer grow, and the software refuses to save. Two answers, to prepare beforehand: archive old financial years to shrink the database, or move to an edition without that limit.
- Our server runs Windows Server 2016: what should we do in 2027?
- Microsoft announces the end of its extended support for 12 January 2027: after that date, no more security fixes under the normal terms. Replacement or an upgrade is therefore prepared now, with a trial migration on a copy. It is the right time to review the database, the disks and the scheduling as well.
- How long does a server last?
- There is no fixed lifespan. The useful markers are the end of the hardware warranty, the system's end-of-support date, and the state of the disks, which fail more often as they age. A server past all three is replaced before the breakdown, not after.
- Is a server better at the office or hosted?
- It depends on your connection, your power cuts, the number of sites and what you want to keep on your premises. A server at the office answers the office workstations quickly; a hosted server serves several sites better and is easier to back up. The guide to hosting details the choice.
- The software is slow at the branch but not at head office: is it the server?
- No. If head office answers in three seconds and the branch in thirty, it is the link between the two, or the way the software crosses that link. A new server would change nothing.